Internal Salesforce Solution Engineering enablement content — a product training walkthrough. Not a login page and not affiliated with or endorsed by Intel Corporation. No account credentials, payment details, or personal information are requested. Maintained by Kou (kou@salesforce.com).
|Headless 360 Workshop
Enterprise Sales·~3 hours·From Call to Quote
Workshop · AI-Assisted Sales
From Inquiry to Execution. Intel AI Accelerates the Workflow.
A raw sales call transcript becomes a CRM-backed quote recommendation —
using MCP servers, Salesforce automation, and the AI tools your team already uses.
One prompt. Zero systems opened.
·
What We're Showcasing
Three Ideas. One Workshop.
Everything we build today maps to one of these three concepts.
AI
AI as a Build & Deploy Partner
Your AI tool understands Salesforce metadata deeply — Flows, Apex, Prompt Templates. Today we use it to deploy artifacts. Soon we'll use it to author them too.
Pillar 1
SF
Salesforce in Your AI Stack
The CRM stops being a system you log into. It becomes a tool your AI reaches for automatically — alongside Google Drive, your inbox, and everything else.
Pillar 2
MCP
Custom MCP = Your Superpowers
You're packaging your discount rules, your approval thresholds, your deal logic into a server any agent can call. You own it. It works with any AI tool.
Pillar 3
·
The Scenario
A $50M Deal. One Sales Call.
Before
The Intel Sales Rep
The rep just finished a call on WB Demo Opportunity and needs to align pricing context, quote history, and next actions. Now what?
×Reconcile notes across IT, finance, AI research, and procurement
×Open Salesforce — check Opportunity, Account, Quote history
×Validate discount thresholds against approval policies
×Build a phased deployment proposal with expansion options
Hours of reconciliation before a quote even starts.
After
The rep opens their AI assistant and types:
"I just got off a call on WB Demo Opportunity. The transcript is in our shared Drive folder. Pull it up, check where we are in Salesforce, and help me build a quote recommendation using the live sandbox record."
+Google Drive (n8n OAuth) — reads and parses the transcript
+Quote Engine — polished recommendation in seconds
Structured quote recommendation. Zero systems opened.
The Output
The Quote Recommendation.
One prompt. Full deal context. Zero systems opened.
QR
Intel Quote Recommendation
WB Demo Opportunity · WB Demo Account
01
Product Bundle Configured
Use the live quote context from WB Demo Quote tied to WB Demo Opportunity. Summarize current line recommendations, pricing posture, and any gaps that still require transcript or product-detail follow-up before final quote submission.
02
Discount Flag VP Approval Required
Validate the requested discount against your configured approval policy using this live record. If policy fields are incomplete, the recommendation must output Data Needed and avoid inventing thresholds or approver titles.
03
Deal Context Strong Signal
Ground context on the actual Salesforce relationships: Opportunity, Account, and linked Quote records. If customer-history or install-base fields are missing, explicitly call that out instead of assuming expansion context.
04
Competitive AMD Watch
Competitive posture should come from transcript text and fields present on the live record. If no competitive intel is captured, return a follow-up question list rather than asserting competitor specifics.
·
Context
What is Headless 360?
It's Salesforce finally saying out loud what's been true since day one: every door is an API. The new bit — agents are first-class clients now.
2000
SOAP API. Day one.
2010
REST API.
2018
MuleSoft.
2024
Agentforce API.
2026
Platform-Hosted MCP.
TDX 2026
26 years of APIs. One brand-new audience: agents.
Layer
Product
What it does
System of Context
Data 360
All your trusted business data — unified, real-time, ready for agents.
System of Work
Customer 360
Decades of business logic and workflows, orchestrated by agents.
System of Agency
Agentforce
Build, deploy, and manage agents at scale across every channel.
System of Engagement
Slack
Where humans and agents come together to get work done.
·
The Build
What You'll Build Today.
Salesforce artifacts + n8n tool connections + a multi-tool AI scenario.
FL
Autolaunched Flow
Triggers on Opportunity stage change, assembles full deal context.
XML
APX
Apex Class
Discount validation, approval threshold logic, quote history lookups.
Apex
PT
Prompt Template
Generates natural language quote recommendation from structured deal data.
Prompt
MCP
n8n MCP Connection
Connect Salesforce in n8n through the native MCP Server with an MCP OAuth2 credential — SSO plus one "Allow access" click, no key file. Google Drive still uses OAuth.
MCP OAuth2
The n8n Workflow Tools
discover
MCP tool call in n8n that takes a natural-language task and finds the relevant Salesforce operation ID.
describe
MCP tool call that returns the technical API schema and required fields for an operation ID.
dispatch / dispatch_readonly
MCP tool calls that execute the request — dispatch_readonly to read, dispatch to change data.
Google Drive (OAuth)
Google Drive connection in n8n used to read and parse call transcripts from shared folders.
These are the native Salesforce MCP Server tools — generic discover → describe → dispatch operations, so the agent
can reach any object without a purpose-built tool per task. They connect through n8n's MCP OAuth2 credential against the MCP Server URL.
Google Drive still uses a normal OAuth credential.
Each attendee sets this up themselves — Intel SSO into the iGPT Agents Platform, then a one-time OAuth "Allow access" click. There is no shared key
file and no consumer key to hand out. Full step-by-step setup is in Connect n8n to Salesforce on the Admin Setup tab.
The surface changes. The platform doesn't.
Salesforce has been headless since the year 2000. Now agents can use all of it.
Intel Royal Silicon Training · Hands-On
Your Hands-On Workshop.
Eight guided modules inside the Royal Silicon app, plus an AI prompt lab. You will click through real screens and change real data — on a record that is yours for the day.
Step 1 · Pick your scenario
Choose your seat tile. Your facilitator gives you a seat number (1–10). Pick it once — every record link and screenshot below will point at your opportunity, so nobody edits anyone else's.
Your recordYou are on . This opportunity and its quote are yours all day — change anything you like.
!You haven't picked a seat yet. The record links below stay locked until you do.
Every module stands on its own. If you fall behind, just jump to the next one — nothing here depends on finishing the module before it.
1
Why Agentic Selling — for Semiconductors
30 min✓ Standalone · discussion only, no app or login needed
Establish the strategic framing for Intel's agentic architecture and operating model.
Why this module comes first
Before you touch a single screen, it helps to know what problem the whole workshop is solving. Selling semiconductors is unusually hard: design cycles run for years, demand swings violently, and a single quote can need sign-off from pricing, finance, and a business unit. Reps lose hours stitching that context together by hand.
An agentic approach flips that. Instead of the rep hunting through systems, an AI assistant gathers the context, checks it against policy, and drafts a recommendation — with a human always making the final call. The rest of the day shows you each piece of that working on real records.
This is a group discussion led from the deck — there is nothing to click in Salesforce. Follow along, and don't worry if you arrive late: the module stands on its own. The two screens below are where the day is heading — the live Opportunity you'll work, and the AI reading it back to you.
Live · sandbox1Where we're going → the live Opportunity
Every module today works a real design-win Opportunity like this one. By the end you'll have shaped its forecast, built its BoM, priced its quote, and had an AI agent read it back to you — hands-on, on your own record.
Live · sandbox1Where we're going → Einstein reads the deal
The agentic payoff: instead of you assembling context, Einstein reads the Opportunity and returns an analysis — the pattern Module 7 turns into a headless agent you just talk to.
Outcome:You can describe, in Intel terms, the target agentic architecture, where the data boundary sits, and who owns what.
1
Frame the business problem
Anchor on semiconductor sales complexity: long design cycles, volatile demand, and cross-functional approvals.
2
Define the agentic pattern
Show the flow: user prompt → context retrieval → policy checks → recommendation → human approval.
3
Map to Intel operating roles
Clarify what Sales, Operations, Finance, and Deal Desk each own in an agentic workflow.
4
State non-negotiable guardrails
Cover audit trail, data boundaries, approval thresholds, and model transparency expectations.
20 min✓ Standalone · discussion only, no app or login needed
Define where Intel can create differentiated value with AI-assisted selling.
Why we map this before building
It's tempting to automate everything at once. The teams that get real value instead pick the two or three moments where AI actually changes the outcome — and ignore the rest. This module is where the room agrees on those moments for Intel, so the hands-on work later feels relevant instead of abstract.
The Opportunity is the spine of the whole day. Reimagining it means asking: what if the record maintained itself — pulling context from documents, email, and market data — so the rep spends time deciding, not typing?
A group discussion from the deck — nothing to click. The two screens below show the "before" (a rich Opportunity record) and the "after" (Einstein Opportunity Update ingesting a document to keep it current) so the reimagined workflow is concrete, not abstract.
Live · sandbox1Opportunity → Summary
Today the Opportunity is where a rep keeps stage, amount, project detail, and next steps current by hand. Reimagining it starts here — with the record everyone already lives in.
Live · sandbox1Opportunity → Einstein Opportunity Update
The reimagined version: paste an email, upload a doc, or forward a thread, and Einstein proposes the field updates. The rep approves — the record maintains itself instead of being retyped.
Outcome:The team prioritizes the top Intel use cases to prototype first and agrees on measurable success criteria.
1
Identify friction points in current selling motion
Capture where reps lose time today: configuration discovery, pricing justification, forecast confidence, and approval latency.
2
Map high-value moments
Rank moments where AI support changes outcomes: design-in, competitive defense, pricing negotiation, and forecast risk.
3
Select the first two lighthouse scenarios
Choose scenarios by business impact, data readiness, and implementation effort.
4
Define outcome KPIs
Set baseline and target for cycle time, conversion, average margin, and approval turnaround.
·
3
Bill of Material
25 min✓ Standalone · start here anytime · needs login + your seat
Build the Bill of Material for your AI-server design win, and see it flow into a quote.
Why the Bill of Material matters
A design win isn't one chip — it's a basket of parts on one board, and the money is in getting that basket right. The BoM Builder pulls in candidate components, tells you which are yours versus a competitor's, and rolls the whole basket up into forecast revenue. That's the difference between quoting one accelerator and quoting the full server content.
In this module you'll add parts, see the "Ours / Sub / No Match" tagging, and generate a quote straight from the board — the same board Module 4 forecasts and Module 5 prices.
Success Check:You build a clean BoM, set socket choices, and generate a quote-ready line set.
First time logging in today?You need to be signed in to Salesforce. The login details are handed out by your facilitator — if you're not in yet, see the login note, then come back. If you land on a page that says "Let's get started!", that is not an error — ignore it and paste the record link below again.
Open your opportunity first.This module works on your seat's opportunity. Pick your seat tile at the top of this page if you haven't, then copy the link below and paste it into your browser.
Your Opportunity · copy
1
Open the Bill Of Material tab
The opportunity sub-tabs are Details · Einstein Market Analysis · Bill Of Material · Forecast · Einstein Opportunity Update · Products · Activities. Click Bill Of Material. The BoM Builder opens: candidate components on the left, your Royal Silicon BoM on the right, with Import BoM CSV, Import BoM PDF and Discover Components across the top.
Live · sandbox1Opportunity → Bill Of Material
Your board shows your own company's name at the top; the layout is identical to this. It already carries a 4-line BoM (UPA, SOM %, Unit ASP, FCST Revenue), so you'll add to it, not start from empty.
2
Discover candidate parts
Click Discover Components to auto-discover, or use Import BoM CSV / PDF. Try at least two paths and watch the Ours / Sub / No Match counters at the top of the candidate panel — confirm the same part is not added twice.
3
Evaluate crosses and add to BOM
Expand a candidate, review the Z2Data grades (A/B/C), filter to Ours, then add selected parts to the BoM panel on the right.
4
Set UPA, SOM %, status, and socket primary
On each BoM line set the line economics (UPA, SOM %) and, in Socket Mode, mark one primary device so alternates do not double-count in the forecast.
5
Generate quote from BOM
Click Generate Quote (top of the opportunity), select the quotable devices, and confirm native quote lines and totals are created. That quote is what Module 5 prices.
Behind, or something looks off?No problem — the next module opens the same opportunity and works whether or not you finished here. Re-paste your opportunity link for a clean start, or move on. Your record is yours all day.
Shift from static reporting to scenario-based forecasting — then shape a real demand curve on your own record.
Why the old way of forecasting breaks
A traditional forecast is one number in a spreadsheet, updated monthly. In semiconductors — where a fab delay or a hyperscaler's design change can move demand 30% in a quarter — that single number is wrong the moment it's saved, and nobody knows by how much.
The shift this module argues for is treating forecast as a decision system with confidence: a baseline plus upside and downside scenarios, each tied to a trigger you can watch for. The second half of the module puts that idea straight into your hands in the Demand Forecast Workbench — you'll spin a simulation beside the baseline and compare the delta, so you're not editing a guess, you're comparing scenarios you can defend.
Outcome:You can explain why a forecast should be managed as a decision system, and you've reshaped a live curve and compared it to baseline.
Part A · The thinking (discussion)
1
Contrast static forecast vs scenario forecast
Show why a single-line forecast fails in volatile supply and demand environments.
2
Introduce confidence-based planning
Discuss baseline, upside, downside scenarios and what operational triggers move between them.
3
Connect forecast to commercial decisions
Tie quantity and timing changes to pricing posture, inventory planning, and approval workflow.
Same record as Module 3 — or start fresh here.If you did Module 3, click the Forecast tab on the opportunity you already have open. If you're jumping straight in, pick your seat tile at the top, copy your opportunity link below, open it, and click Forecast.
Your Opportunity · copy
4
Open the Demand Forecast Workbench
Click the Forecast tab. Confirm the header reads the right Opportunity, Configuration (AI Training Server — Auto Config), Scenario (Baseline), and SKU Shares (4 sockets with competitive share) before you touch anything.
Live · sandbox1Opportunity → Forecast Workbench
Your opportunity shows your own company's name; the workbench is identical. It's pre-seeded to 36 months / 480,000 units from a 1 Apr 2027 production start — a ramp you can reshape and compare, not a blank form.
5
Pick the model that fits the ramp
Open the model gallery and compare shapes — Bass, Gompertz, Holt-Winters, Moirai and more. Different end-markets ramp differently; the model you pick is the starting opinion you'll then curate.
Live · sandbox1Forecast → Model Gallery
Eight model shapes side by side. Each is keyed to how a given application ramps — a consumer part and a data-center accelerator do not climb the same way, and the gallery lets you choose rather than assume.
6
Create a simulation and reshape with AI
Click New Sim, then adjust periods in the grid or drag the chart. Try a plain-language reshape — for example, "push the peak out by 3 months" — and watch the curve redraw.
7
Compare and publish
Click Compare vs Baseline, read the revenue and quantity delta, then publish to the Opportunity Product Schedule if you're happy with the scenario.
Ran long? Skip ahead freely.Module 5 opens a different tab and doesn't need anything you did here. Re-paste your opportunity link if the page got into a weird state, or just move on — your record keeps whatever you saved.
·
5
Using Data Science for Quoting
45 min✓ Standalone · start here anytime · needs login + your seat
Read the pricing signals on a real quote, then make one defensible, line-by-line pricing decision.
Why price a quote this way
Left alone, reps price from gut feel and a stale spreadsheet — and margin leaks on every deal. The Quote Pricing Workbench puts a floor, a ceiling, and a market-index reference on every single line, so the rep can see at a glance whether a price is inside policy or needs a sign-off. No more guessing where the guardrails are.
The reason to work line-by-line is that each product has its own floor, ceiling, and market signal — a blanket "10% off the quote" hides which lines are healthy and which are bleeding. Here you open one line, read its Demand Outlook and Swappability score, and either price within the band or submit an override with a reason an approver can act on. That override note is the quiet hero: it's the audit trail that lets Deal Desk say yes fast.
Success Check:You produce one quote-line decision with evidence: the index signal, the competitive context, and your override call with a rationale.
First time logging in today?You need to be signed in to Salesforce. If you land on a page that says "Let's get started!", that is not an error — ignore it and paste the record link below again.
Open your quote first.This module works on your seat's quote. Pick your seat tile at the top of the page, then copy the link below and paste it into your browser's address bar.
Your Quote · copy
1
Open the Pricing Workbench tab
On the quote, look at the row of tabs under the quote name: Details · Quote Line Editor · Pricing Workbench · Competitive Analysis · Approvals · Pricing Schedule. Click Pricing Workbench. You'll see Quote Pricing Workbench v2 with one row per product — each showing Quantity, Approval status, and Net/Total price.
Live · sandbox1Quote → Pricing Workbench
Your quote shows your own company's name at the top, but the layout and product lines are identical. Every line reads InRange here — the in-policy state. When a line falls outside the band, its Approval column changes, which is what step 4 has you drive.
2
Open a line workspace and read the boundaries
Click a product name to open its line workspace. Read Floor Price, Ceiling Price, Weighted Index Price and the Pricing Outcome. If the net price sits inside floor–ceiling the line is in-policy; below floor it needs escalation. Say, out loud, which state the line is in and why.
Live · sandbox1Pricing Workbench → Line Detail + AI Advisor
Inside a line: the Index Baseline sets the floor/ceiling band, and Ask the Pricing Advisor explains the recommendation in plain language. This is the "data science" made legible — the numbers argue, you decide.
3
Use market outlook and swappability
In the intelligence panel, read Demand Outlook and the Swappability score. Use them to argue a posture — a low-swappability part with strong demand can hold price. Cross-check with the Competitive Analysis tab if you want the head-to-head.
4
Execute the pricing action with a rationale
Save within-band pricing, or submit a Pricing Override with a reason and an approver-ready note. State the expected margin impact out loud before you save — that note is what an approver reads and what enablement replays later.
No harm done if you experiment.This is your own quote — override a price, change your mind, override it back. Nothing you do here affects another trainee. The next module reads a different set of records and works independently of anything you saved here.
See how the Business Unit builds forecast indexes and Finance builds pricing indexes — and how each one steers the workbenches you just used.
Why indexes are the hidden engine
In Modules 4 and 5 you saw a demand curve and a price floor. Neither is arbitrary — both are pinned to market indexes that different teams own. The Business Unit curates forecast indexes (demand signals like DRAM/NAND spot trends) that shape the demand curve; Finance curates pricing indexes (cost and market references) that set the floor and ceiling on every quote line.
The Index Workbench is where those indexes are built, assigned, and — crucially — self-tuned from real win/loss feedback. When win rate drifts below target, the workbench recommends an adjustment. This module follows one index from its source data, through its KPIs, to the recommendation that flows back into pricing.
Success Check:You can explain how a BU forecast index shapes the demand curve and how a Finance pricing index sets a quote-line floor — and read the self-tuning recommendation for one index.
Shared reference data — read, don't overwrite.The Index Workbench reads indexes the whole room shares. Explore freely, but treat the indexes as reference data. Open it from the App Launcher → Index Workbench, or use the link below.
Index Workbench · copyhttps://trailsignup-53653ba61aae20--sandbox1.sandbox.lightning.force.com/lightning/n/Index_Workbench
How the BU & Finance build indexes
1
Open the Index Workbench and read the index list
On the left is the library of indexes — memory spot (DRAM DDR5, NAND Flash), commodity grades, and strategic references like the PwC AI-server price trend. Each shows its symbol, point count, and how many quote lines (scopes) it's assigned to. This is the shared brain both the forecast and pricing engines read from.
Live · sandbox1Index Workbench → Library & Ingest
The index library (left) and the ways to feed it (right): paste a CSV, blend several indexes, let Einstein extract a price series from a report, or web-scrape. This is how BU and Finance keep the signals current without a data-engineering ticket.
2
Select an index and read its KPIs & self-tuning recommendation
Click an index (for example DRAM DDR5 Spot). The right panel shows its KPIs against real deals — Win Rate vs target, Price Realization, Market-Share Signal, and Deals Tied — then a Recommended Adjustment computed from won/lost feedback.
Live · sandbox1Index Workbench → KPIs & Recommended Adjustment
Here win rate is 40% against a 60% target and market share is slipping, so the workbench recommends a −14.79% index adjustment — the exact self-tuning loop that flows back into the pricing floor you saw in Module 5.
How it flows back into the workbenches
3
Forecast side — BU indexes shape the demand curve
A forecast index (a demand signal) biases the model in the Demand Forecast Workbench: when the BU's spot-demand index turns up, the ramp the model proposes in Module 4 shifts with it. The BU curates the signal once; every opportunity on that application inherits it.
4
Pricing side — Finance indexes set the floor
A pricing index sets the Floor / Ceiling / Weighted Index Price you read on a quote line in Module 5. When Finance updates the index — or accepts the self-tuning recommendation — the guardrails on every scoped quote line move with it, so contract pricing stays fair as the market shifts.
5
Propose one adjustment with a rationale
Given the win rate against target and the price trend, argue for one index adjustment. State the expected win/margin tradeoff and what evidence would prove you wrong — then note that the workbench already computed a recommendation you can weigh against your instinct.
Rule of thumb:Weigh win-rate elasticity and loss reasons together — don't cut an index just because a few deals were lost. A deal lost on delivery time shouldn't drag down price.
·
7
AI Prompt Lab — Salesforce as a Tool (n8n)
20 min✓ Standalone · runs in n8n · no Salesforce login needed
Watch an AI agent reach for Salesforce as a tool it decides to call — the whole point of Headless 360.
Why this is the "aha" moment
Everything so far had you drive Salesforce. Now you'll do the opposite: you type a plain-English question into a chat, and an AI agent decides on its own to call Salesforce to answer it — pulling live records, reasoning over them, and writing back a clean answer. You never open the app. That's "Headless 360": Salesforce as a tool an agent reaches for, not a place a human logs into.
These ten prompts are deliberately bounded — each names a specific record and tells the agent to stop if data is missing. That's what keeps the agent fast and stops it spinning in circles, which is the number-one way these demos go wrong.
These ten prompts run in the n8n AI Agent node with the Salesforce node attached as a tool. The agent reads live sandbox1 records and reasons over them; you never open Salesforce. Every prompt is deliberately bounded so the agent answers in a pass or two rather than looping.
Runs in n8n.Use the MCP OAuth2 credential you connected in Connect n8n to Salesforce on the Admin Setup tab, then paste a prompt into the AI Agent's chat. Once that one-time OAuth connection is done, you do not sign in again for this module.
Why these ten won't hang the agent.An n8n AI Agent loops tool calls until it answers or hits Max Iterations (default ~10) — then it stops with "Agent stopped due to max iterations." Two things cause that loop: (1) an unbounded query that pulls hundreds of records into context, and (2) hunting for data that isn't there. Every prompt below names specific records (never "all opportunities") and ends with an explicit "if a field is blank, say Data Needed and stop." That keeps each run fast and cheap on tokens. Set the agent's Max Iterations to 5 as a backstop.
Canonical record set (verified in sandbox1, 3 Aug 2026).Opportunities: WB Demo Opportunity, RS P6 E2E Opp, Ship and Debit Demo, Omega, Inc. - New Business - 40K, Larry Baxter - Add-On Business - 6K. Linked quotes: WB Demo Quote, RS P6 E2E Quote, Ship and Debit Demo- Quote -2026-04-07, United Partners, Baxter Proposal. Known-true data shape: WB Demo and Ship and Debit have blank Amount — that is intentional, and prompts 1 and 7 turn it into the point. RS P6's quote carries clean 10% and 20% line discounts, so prompt 2 does real math. Omega has a rich Description but no named competitor, so prompt 6 returns discovery questions instead of inventing one.
Success Check:Each participant runs one prompt as-is and watches the agent call the Salesforce tool and answer from live data, then tunes one constraint (a record name, a format, a threshold) and re-runs to see the output change.
Ten ready-to-run agent prompts
Deal Brief
Create a one-screen deal brief for WB Demo Opportunity. First use the Salesforce tool to get the Opportunity and its linked Account and Quote. Report: stage, amount, quote grand total, top 3 risks, and the single next best action for this week. If a field such as Amount is blank, list it under "Data Needed" and stop — do not guess and do not keep searching.
Pricing / Approval
For RS P6 E2E Opp and its RS P6 E2E Quote, use the Salesforce tool to read the quote line items. For each line compute the discount as (List - Unit) / List. State whether a 7% discount request is inside the discount already granted, name the likely approver role, and write a two-sentence approval justification I can paste into Salesforce. If approval-threshold fields are not present, say "Data Needed" for those and continue with what you have.
Quote Reconciliation
Compare the WB Demo Quote grand total against the sum of its quote line item totals using the Salesforce tool. Report both numbers, the difference, and whether they reconcile. Keep the answer to 4 lines.
This-Month Close Watch
Using the Salesforce tool, get these five opportunities only: WB Demo Opportunity, RS P6 E2E Opp, Ship and Debit Demo, "Omega, Inc. - New Business - 40K", "Larry Baxter - Add-On Business - 6K". List any whose CloseDate is on or before 2026-08-31 and is not already Closed. For each, give name, stage, close date, and whether it looks at risk. Do not pull any other opportunities.
Overdue Triage
From the same five canonical opportunities, use the Salesforce tool to find any with a CloseDate in the past relative to 2026-08-03 that are still open. For each, state how many days overdue and one corrective action. If none are overdue, say so plainly.
Competitive Brief
Build a competitive response brief for "Omega, Inc. - New Business - 40K". Use the Salesforce tool to read the Opportunity Description, NextStep, and Account industry/type. Base competitor points ONLY on what is written in those fields. If no competitor is named in the record, return a list of discovery questions to ask instead of naming a competitor. Do not read the linked quote — it has no line items.
Data-Quality Audit
Audit data quality for WB Demo Opportunity and Ship and Debit Demo using the Salesforce tool. For each, list which of these fields are blank: Amount, Type, LeadSource, NextStep, Description. Explain in one line why each blank matters for a pricing or forecast decision. Do not attempt to fill the values.
Account 360
Give a 5-line account summary for the account behind "Omega, Inc. - New Business - 40K". Use the Salesforce tool to read Account Name, Industry, Type, and AnnualRevenue, plus the count of related open opportunities. If AnnualRevenue is blank, say "not recorded".
Strict JSON
Return the result for RS P6 E2E Opp as strict JSON only — no prose, no markdown — with keys: opportunityName, stage, amount, quoteGrandTotal, topRisk, nextAction. Use the Salesforce tool to fill each value. If a value is unavailable put null. Keep topRisk and nextAction under 12 words each.
Two-Deal Compare
Compare RS P6 E2E Opp and "Larry Baxter - Add-On Business - 6K" using the Salesforce tool. In a 2-row table give: opportunity, stage, amount, close date, and which is closer to closing and why. Read only these two opportunities.
These are read/reason prompts, not build prompts.The n8n AI Agent returns text or JSON into a chat — it cannot draw a Kanban board or generate a Salesforce component. Anything shaped like "generate a UI / build a board / create a tab" belongs on the Coding Prompts tab (bottom toolbar), where an AI coding assistant writes and deploys the real component. That is the deliberate escalation: here Salesforce is a tool the agent reads; on the Coding Prompts tab it becomes a target the agent builds into.
·
8
Using Prompts for Vibe Coding
15 min✓ Standalone · runs in n8n · no Salesforce login needed
Take the six core patterns and adapt one to a deal you care about.
Why learn the shape, not just copy prompts
Running someone else's prompt is a party trick; writing your own is the skill. Every reliable agent prompt has the same two habits baked in: name the exact record (so the agent doesn't wander), and tell it to stop and say "Data Needed" if a field is blank (so it doesn't loop forever hunting). Learn those two habits here and you can point the agent at any deal on your desk without it hanging or hallucinating.
Module 7 gave you ten prompts to run. This module is about the shape behind them so you can write your own without tripping the agent. Each pattern below is copy-paste ready and follows the same two safety rules: name the record, bound the output.
Runs in n8n — same credential as Module 7.Same MCP OAuth2 credential. Click Copy, paste into the AI Agent chat, change only the record name or the constraint you want to test.
How to use:Copy a pattern, swap in one of the five canonical opportunities, and keep the closing "if blank, say Data Needed and stop" clause — that clause is what keeps the agent from looping.
Deal Brief
Deal Brief
Create a one-screen deal brief for <OPPORTUNITY NAME>. Use the Salesforce tool to get the Opportunity plus its linked Account and Quote. Report stage, amount, quote grand total, top 3 risks, and the single next best action this week. If a field is blank, list it under Data Needed and stop — do not guess or keep searching.
Pricing / Approval
Pricing / Approval
For <OPPORTUNITY NAME> and its linked quote, use the Salesforce tool to read the quote line items and compute each discount as (List - Unit) / List. Say whether a <N>% discount is inside the discount already granted, name the likely approver role, and write a two-sentence approval note. If threshold fields are missing, say Data Needed and continue.
Close-Date Watch
Close-Date Watch
Using the Salesforce tool, get only these named opportunities: <LIST 2-5 NAMES>. List any whose CloseDate is on or before <DATE> and not already Closed, with stage and whether it looks at risk. Do not pull any other opportunities.
Competitive (No Invention)
Competitive
Build a competitive brief for <OPPORTUNITY NAME>. Use the Salesforce tool to read the Opportunity Description, NextStep, and Account industry/type. Base competitor points ONLY on what is written there. If no competitor is named, return discovery questions instead of naming one.
Data-Quality Audit
Data Quality
Audit data quality for <ONE OR TWO OPPORTUNITY NAMES> using the Salesforce tool. List which of Amount, Type, LeadSource, NextStep, Description are blank, and one line each on why the blank matters for a pricing or forecast decision. Do not fill the values.
Strict JSON Output
Output Formatting
Return the result for <OPPORTUNITY NAME> as strict JSON only — no prose, no markdown — with keys opportunityName, stage, amount, quoteGrandTotal, topRisk, nextAction. Use the Salesforce tool to fill each value; use null if unavailable. Keep topRisk and nextAction under 12 words.
Facilitator Note:Have each table pick one pattern, adapt it to a different canonical opportunity, and read the agent's answer aloud. Then break one on purpose — remove the "if blank, stop" clause and point at a sparse record — to show why the guardrail clause matters.
From strategy to execution. Built for Intel teams.
Everything the facilitator does before the room arrives — logins, the shared account, MFA reality, and the deploy tooling — plus the one self-service step each attendee runs to connect n8n to Salesforce for Modules 9 and 10.
!
Before You Start: Access & Login
Read first · 5 min
How to get into sandbox1 — and the one screen that looks like an error but isn't.
Everything below was built and verified in sandbox1, most recently re-walked end to end on 3 August 2026. Modules 5–8 need a Salesforce login to work in the app. Modules 9 and 10 run in n8n. Modules 11 and 12 also need a login — you deploy what you generate. This login base is the one to copy: https://test.salesforce.com.
!
The first screen after you log in is NOT an error. Do not click "Sign in with Email".
Immediately after a successful login you will land on a page that looks like this:
Let's get started!
Sign in with Email
This is a leftover trial-org landing page. Your session is already valid and you are already logged in. Nothing has failed.
×Do not click "Sign in with Email." It opens a First Name / Last Name registration wizard that is a dead end — you will have to back out and start again.
✓Do this instead: ignore the page completely and navigate to the app — use the App Launcher (the grid icon, top left) or the direct app URL your facilitator puts on screen.
Attendee Login — Modules 5 to 8 and 11 to 12
One shared account, verified running the Bill of Material, Forecast, Pricing Workbench, and Competitive Analysis surfaces end to end against the Intel Gaudi 3 flagship record on 3 August 2026. It also carries the Intel_Workshop_Deploy permission set so you can deploy what you build in the Vibe Coding modules (11 and 12). A password login was re-tested after that elevation and still works: password only — no passkey, no MFA challenge.
Shared Attendee Account
Login URLhttps://test.salesforce.com
Usernameintel.workshop.user@sandbox1.workshop
Password— provided by your facilitator —
Verified runningPricing Workbench · BOM Builder
Which Modules Need Salesforce?
Modules 1–4No — strategy and discussion
Modules 5–8Yes — shared attendee login (in-app)
Modules 9–10No — n8n only
Modules 11–12Yes — generate in Cursor, deploy to the org
On the password:The live password is deliberately not printed in this document, because this document gets shared. Your facilitator will read it out or pin it in chat on the day.
Why You Might Hear About Passkeys
Context, not a task:This org sits in Salesforce's July 2026 MFA enforcement wave. Privileged users — anyone on the System Administrator profile, or with Modify All Data, View All Data, Customize Application, or Author Apex — are forced into phishing-resistant passkey enrollment with no alternative method offered. On locked-down Intel Windows laptops, Windows Hello is unavailable, so passkey enrollment falls back to a QR-code cross-device flow that requires a second device. That is unworkable for a room of twenty. Customize Application is one of those four permissions, and the attendee account now holds it via Intel_Workshop_Deploy so that attendees can deploy — which makes the account privileged by Salesforce's definition. A password login was tested after that elevation and no passkey wall appeared: login succeeds, LoginHistory records Success, and the session is valid. So in practice you still sign in with a password. If Salesforce tightens privileged-user enforcement on this instance, this account is now in scope, and the fallback is the JWT frontdoor path below — it never renders a login page. If you do land on a passkey screen, stop and tell the facilitator.
Connect n8n to Salesforce — Modules 9 and 10
Modules 9 and 10 run in Intel's own n8n (the iGPT Agents Platform) talking to Salesforce through Salesforce's native MCP Server. Each attendee does this once, themselves — it is the only self-service setup on this page. It has three parts that hand values back and forth: get into n8n, turn on the Salesforce MCP Server + External Client App, and wire the two together with an OAuth handshake. Follow the Intel demo video alongside these steps if you get stuck.
Why this replaces the old key-file method:There is nothing to hand out — no consumer key, no server.key. You authenticate through Intel SSO and a normal OAuth "Allow access" click. The Salesforce MCP Server exposes generic discover / describe / dispatch tools, so the agent can read (and, with the right server, write) any object without a purpose-built tool per task.
Part 1 · Get n8n access (iGPT Agents Platform)
1
Open iGPT and provision access on the fly
Go to https://igpt.intel.com/, click Agents in the lower left, then click the Agents Platform link in step 1. That SSO link provisions your n8n access automatically — use the SSO link, not a username/password. You will arrive at a blank page in n8n; that is expected.
Wait for your token budget.A $10 monthly inference budget is provisioned within about 10 minutes. Refresh the iGPT Agents page to see it appear. You can build the workflow before it lands, but the agent will not answer until it does.
Part 2 · Salesforce — enable the MCP Server and create the External Client App
2
Enable the MCP Server and copy its URL
In Salesforce Setup, open MCP Server and enable the server you want to interact with. Copy the Server URL — this is the endpoint your n8n MCP client will call. Browse the tools listed inside it to see what that server can do.
Headless 360The full read/write server exposing discover, describe, dispatch, and dispatch_readonly
3
Create an External Client App (ECA)
Setup → External Client Apps → New. Enter basic info (app name, e.g. n8n MCP Client), then Enable OAuth. Leave the Callback URL empty for now — n8n gives you that value in Part 3, and you will come back to paste it in. Continue to the OAuth settings below.
Click Create, then copy the Client ID and Client Secret — you paste these into n8n in Part 3. The credentials can take up to ~10 minutes to become active after creation.
Disable PKCE at the org level too.In Setup → OAuth and OpenID Connect Settings, disable PKCE org-wide. The ECA-level uncheck is not sufficient on its own — the org setting must also be off or the OAuth handshake in Part 3 fails.
Part 3 · n8n — build the workflow and connect the credential
5
Build a minimal workflow
Create your first workflow and choose Start from scratch. Add On chat message as the first step (defaults are fine — just close the window). Then add AI → AI Agent as the next step.
6
Set the AI Agent options
If you are using an sobject MCP server, all defaults are fine. If you are using the headless-360 MCP server, add a System Message option (below) so the agent uses the discover→describe→prepare→dispatch sequence. Set the Chat model and Memory as listed.
System Message — headless-360 only
You are a Salesforce administrative assistant. You have access to the Salesforce Headless 360 MCP tools.
When asked to perform a Salesforce task, you MUST use the tools autonomously in this exact sequence:
1. DISCOVER: Call the `discover` tool with a natural language description of the task to find the relevant operation ID.
2. DESCRIBE: Call the `describe` tool using the operation ID from step 1 to retrieve the technical API schema and required fields.
3. PREPARE: If you need to look up IDs (like a ProfileId, PermissionSetId, or existing user details), use `discover` and `dispatch_readonly` to query Salesforce for that prerequisite data.
4. DISPATCH: Formulate the final request using the schema from step 2 and the data from step 3. Use `dispatch` to execute data changes, or `dispatch_readonly` for read operations.
Use Claude Sonnet 4.5 — not the low-cost model.The low-cost model (Llama 3.x) does not drive the tool sequence properly. Sonnet 4.5 works and is cheaper to run than 4.6, so it is the right default for this workshop.
7
Add the MCP Client Tool and its credential
Add a Tool → MCP Client Tool to the agent. Set the Endpoint to the Server URL you copied in Part 2, choose MCP OAuth2 authentication, and create a new credential. n8n will show an OAuth Redirect URL — copy it and paste it back into your ECA's Callback URL in Salesforce (Part 2, step 3).
Client ID & SecretPaste from the ECA (Part 2, step 4)
8
Connect and authorize
Click Connect, go through the OAuth flow, and Allow access. Close the window when it finishes. Your credential is now live.
Prove it works
Send this prompt in the n8n chat. It works against any org and should cost only a few cents of inference budget.
Smoke test — paste into n8n chat
List the top 10 users in the system, sorted by username. I only need user first name, last name, username, and email address.
Nothing to hand out, nothing secret in this page.The Client ID and Secret live only in your own n8n credential, and access is gated by Intel SSO plus the OAuth "Allow access" click. There is no shared key file for this flow.
Facilitator Only
Facilitators do not use the shared attendee login. Use the admin account below and open it from the CLI — sf org open issues a frontdoor URL that bypasses the passkey wall entirely.
Facilitator Access
Usernameintel.workshop@sandbox1.com
Org aliassandbox1
How to get inCLI frontdoor URL — command below
Facilitator — open the org
sf org open --target-org sandbox1
Use this session for all admin Setup work.Attendees deploy their own module 11 and 12 output from the CLI under Intel_Workshop_Deploy, so you no longer have to deploy on their behalf. They still have no admin Setup access and do not need any.
Deploy rights are granted — Modify All Data and View All Data are not:The attendee account holds Intel_Workshop_Deploy (Customize Application, Author Apex, Modify Metadata plus required dependencies) because metadata deployment is impossible without them — an LWC bundle deploy fails with INSUFFICIENT_ACCESS on LightningComponentBundle until Customize Application is granted. Leave Modify All Data and View All Data off: View All Data would override the region-scoped visibility that RS_Region_Field_Access provides and distort the Forecast and Pricing modules. This was a permission-set grant, not a profile switch — the account stays on SDO-Sales, so demo page layouts and record-type defaults are preserved.
Do not refresh the sandbox before the workshop:A refresh resets MFA verifiers and can disturb the login configuration this plan depends on. If a refresh is unavoidable, do it with enough lead time to re-verify every login in this section.
Pre-Workshop Checklist — Run on the Morning
1
Verify the shared attendee login works
Sign in as intel.workshop.user@sandbox1.workshop from an incognito window and confirm it still reaches the app with a password only. This was verified working after the Intel_Workshop_Deploy elevation, with no passkey prompt — but the account is now privileged by Salesforce's definition, so this smoke test matters more than it did before. Run it on a real Intel Windows laptop if you can: the verification browser exposed no platform authenticator, and a machine with Windows Hello available might be offered a passkey. If a passkey screen appears with no way past it, fall back to the CLI frontdoor path above.
2
Confirm the n8n MCP credential connects
Open the MCP OAuth2 credential in n8n (the one wired to the Salesforce MCP Server), run its connection test, then execute one workflow that returns a live record — the smoke-test prompt is ideal. A green credential test alone is not enough: prove a tool call returns data. If the OAuth handshake fails, re-check that PKCE is disabled both on the ECA and at the org level.
3
Confirm the hands-on records load
Open the Intel Gaudi 3 flagship opportunity and quote (URLs are printed at the top of every hands-on module, 3–6) and confirm the Bill of Material, Forecast, Pricing Workbench, and Competitive Analysis tabs all render with data. Also confirm the five n8n record-set opportunities in Module 7 still exist.
4
Have the direct app URL on screen
So the room can skip the "Let's get started!" landing page entirely rather than twenty people hitting it at the same moment. Each hands-on module also carries its own copy-paste record URL, so an attendee who lands on the interstitial can paste straight past it.
5
Distribute the JWT key and consumer key for the Cursor deploy setup
Only needed if you run the optional Coding Prompts track. Those attendees run the JWT login in Set Up Cursor to Deploy (below). Have the server.key file and the connected app consumer key ready to hand out — neither is in this document. Confirm the connected app still has "Issue JSON Web Token (JWT)-based access tokens" switched off: with it on, metadata deploys fail with "SOAP API does not support JWT-based access tokens." Run the JWT login once yourself on the morning to prove the key still authorises, and remind the room that sf org login web does not work against this org.
·
!
Set Up Cursor to Deploy: MCP & Org Connection
Do this before Module 8 · 10 min
Modules 11 and 12 ask you to deploy what you generate. This is the one-time setup that makes that possible.
You are adding a second MCP server to Cursor. An MCP server is simply a helper that gives Cursor extra abilities. The one already configured can only read data from Salesforce. The new one — Salesforce's official DX MCP server — can deploy metadata, which is what modules 11 and 12 need. Both are locked to sandbox1 and can reach no other org. Verified end to end in sandbox1 on 2 August 2026.
Do not use sf org login web for this org.It was tested three times, including from a clean browser profile, and it never completes. The password is accepted, but the org then redirects to the same "Let's get started!" page described above instead of returning to Cursor — so the command sits there waiting forever. Use the JWT command in step 2 instead. It needs no browser at all and finishes in about five seconds.
Setup — Five Steps
1
Check you have the two tools installed
Run both commands below. If either one reports "not recognized" or "command not found", tell your facilitator before going further — nothing after this step will work without them.
Step 1 — check prerequisites
node --version
sf --version
2
Connect Cursor to sandbox1
Your facilitator will give you two things: a key file (server.key) to save on your machine, and a consumer key (a long string of letters and numbers). Put them into the command below in place of the two bracketed placeholders, then run it. Keep the alias exactly as sandbox1 — the MCP config in step 4 looks for that exact name.
You should see Successfully authorized intel.workshop.user@sandbox1.workshop. Confirm it stuck:
Step 2 — confirm the connection
sf org display --target-org sandbox1
3
Create the folder you will work in
The deploy tool only works inside a Salesforce project folder — a folder containing a file called sfdx-project.json. Without it the deploy fails, so do not skip this. Run the commands below, then open that intel-vibe folder in Cursor.
Step 3 — create your project folder
sf project generate --name intel-vibe --template empty
cd intel-vibe
4
Add the MCP configuration
Inside the intel-vibe folder, create a new file named exactly .mcp.json (the leading dot matters) and paste the block below into it. Save the file. Change nothing — there is nothing in here you need to personalise.
Quit Cursor completely and reopen the intel-vibe folder. The very first start downloads the server and can take up to a minute with no visible progress — that is normal, let it finish. Then ask Cursor the question below. If it answers with the org details, you are ready for modules 11 and 12.
Step 5 — ask Cursor this
Using the Salesforce MCP server, which org am I connected to? Then run this SOQL against sandbox1: SELECT Id, Name FROM Account LIMIT 3
What You Just Enabled
Tools Cursor Can Now Use — sandbox1 Only
deploy_metadataPushes your generated LWC and tab into sandbox1. Asks your permission every time.
retrieve_metadataPulls existing metadata out of the org into your folder. Asks your permission.
run_soql_queryReads records. Runs without asking — it cannot change anything.
run_apex_testRuns Apex tests. Asks your permission. Not needed for modules 11 or 12.
get_usernameTells Cursor which org to target. Runs without asking.
Cursor will ask before it deploys — that is deliberate.Only the two read-only tools are pre-approved. Anything that writes to the org stops and waits for you to click approve, so you always see what is about to be pushed into a sandbox nineteen other people are sharing. Read the component list in the prompt before you approve it.
No secrets are in this file.The config above contains no password, key, or token, which is why it is safe to share. Your connection lives outside it, created by the command in step 2. The server.key file your facilitator hands out is sensitive — keep it on your own machine and delete it after the workshop.
Deploys queue for the whole room.Metadata deployments run one at a time across the entire org at roughly 40 seconds each. If twenty people deploy at the same moment the last one waits about thirteen minutes. Deploy when you are ready rather than all together, and if a deploy seems slow it is almost certainly queued behind someone else — let it run rather than cancelling and retrying.
Give everyone the same component name prefix.Everyone deploys into one shared org as one shared user, so two people using the same component name will overwrite each other's work with no warning. Put your initials in front of every component you create — abcDealQueue, not dealQueue.
If step 2 fails on the day:Tell your facilitator rather than trying sf org login web — that path is known not to work here. The fallback is a one-line authorisation file the facilitator can generate and hand over, used with sf org login sfdx-url. If your own connection cannot be fixed quickly, pair with someone whose setup works; the modules run fine with two people at one screen.
·
Optional · After the Workshop
Set Up for Coding.
The workshop itself needs none of this. This page is for anyone who wants to go further and have an AI coding assistant (Cursor or Claude Code) build and deploy new Royal Silicon workbenches straight into the sandbox.
Why this is a separate page
In the workshop you saw Salesforce being read by an AI agent (the n8n prompt lab). This page is the next step up: an AI coding assistant that writes and deploys real components into the org. That is a bigger privilege — it can change metadata — so it lives here, apart from the trainee workshop, and is entirely optional.
The mechanic is simple. You connect your coding assistant to the sandbox once, then you describe the workbench you want in plain English. The assistant generates the Lightning Web Component, a tab, and the metadata, then deploys it and adds it to the Royal Silicon app — all from a prompt.
A
Connect Your Coding Assistant
One-time · 10 min
The full, step-by-step connection (MCP config, JWT login, the folder you work in) lives on the Admin Setup tab under “Set Up Cursor to Deploy”.
Do the Admin Setup steps first.Open the Admin Setup tab, complete “Set Up Cursor to Deploy: MCP & Org Connection” (five steps), then come back here. When Cursor can answer “which org am I connected to?” with sandbox1, you are ready.
One shared org, one shared login.Everything you deploy lands in the same sandbox1 that the whole room shares. Prefix every component name with your initials (for example abcMarginWorkbench) so you never overwrite someone else's work, and deploy when you're ready rather than all at once — deploys run one at a time across the org.
Why these six prompts
Each prompt below builds a brand-new workbench that does not exist in Royal Silicon yet, then adds it to the app as its own tab. They are written to be self-contained: the assistant creates the LWC, the .js-meta.xml, and a CustomTab, deploys through the Salesforce DX MCP, and adds the tab to the Royal Silicon Lightning app. Read the generated code before you approve the deploy — the point is to see how far a plain-English request gets you, not to trust it blindly.
Pick one and run it end to end. If you have time, try a second. Each is independent — you do not need to run them in order.
B
Six Workbench-Builder Prompts
Pick one · ~15 min each
Copy a prompt into Cursor or Claude Code, replace abc with your initials, and let it build, deploy, and surface the tab.
Six ready-to-run builder prompts
1 · Margin Guard Workbench
Build a Salesforce Lightning Web Component called abcMarginGuard (replace abc with my initials). It shows, for a chosen Quote, every quote line item with its List price, Net price, discount %, and a computed gross-margin % against a 30% floor — flagging any line below floor in red. Include a KPI header with count of below-floor lines and total at-risk revenue. Generate the HTML, JS controller, the js-meta.xml exposing it to a Quote record page, and a CustomTab named abcMarginGuard. Then deploy everything to the sandbox1 org using the Salesforce DX MCP, and add the tab to the Royal Silicon Lightning app. Use a SOQL query on QuoteLineItem for live data; do not hardcode record IDs.
2 · Design-Win Scoreboard
Build a Salesforce Lightning Web Component called abcDesignWinBoard (replace abc with my initials) that lists Design_Registration__c records grouped by Status__c into columns (Submitted, Approved, Rejected), each card showing Design_Name__c, End_Customer_Account__r.Name, and Total_Assemblage_Quantity__c. Add a KPI header with total approved quantity. Generate HTML, JS, js-meta.xml for an app page, and a CustomTab named abcDesignWinBoard. Deploy to sandbox1 via the Salesforce DX MCP and add the tab to the Royal Silicon app. Query live data with SOQL; no hardcoded IDs.
3 · Forecast Confidence Panel
Build a Salesforce Lightning Web Component called abcForecastConfidence (replace abc with my initials) for an Opportunity record page. It reads the Opportunity fields Forecast_Confidence_Score__c, Historical_Accuracy_Score__c, Risk_Score__c and Assemblage_Forecast_Units__c and renders them as four labelled gauges with a one-line plain-English interpretation of each. Generate HTML, JS, js-meta.xml (target: Opportunity record page), and a CustomTab named abcForecastConfidence. Deploy to sandbox1 through the Salesforce DX MCP and add it to the Royal Silicon app. Use getRecord or SOQL for live values; no hardcoded IDs.
4 · Competitive Cross Table
Build a Salesforce Lightning Web Component called abcCompareGrid (replace abc with my initials) that, for a chosen Opportunity, shows its OpportunityLineItems as rows with columns Product, Units Per Assemblage, SKU Share %, Unit Price, and extended value (quantity x unit price). Add a totals row. Generate HTML, JS, js-meta.xml for a Quote or Opportunity page, and a CustomTab named abcCompareGrid. Deploy to sandbox1 via the Salesforce DX MCP and add the tab to the Royal Silicon app. Live SOQL only; no hardcoded IDs.
5 · Approval Queue Workbench
Build a Salesforce Lightning Web Component called abcApprovalQueue (replace abc with my initials) that lists QuoteLineItem records where Pricing_Approval__c = 'Pending', showing Quote name, Product, discount %, and the approver tier. Add a KPI header counting pending approvals and summing their total price. Generate HTML, JS, js-meta.xml for an app page, and a CustomTab named abcApprovalQueue. Deploy to sandbox1 through the Salesforce DX MCP and add the tab to the Royal Silicon app. Query live data with SOQL; do not hardcode IDs.
6 · Account AI Server Portfolio
Build a Salesforce Lightning Web Component called abcAccountPortfolio (replace abc with my initials) for an Account record page. It lists that Account's Opportunities with Name, StageName, Amount, and CloseDate, plus a KPI header of total open pipeline and count of AI-server design wins. Generate HTML, JS, js-meta.xml (target: Account record page), and a CustomTab named abcAccountPortfolio. Deploy to sandbox1 via the Salesforce DX MCP and add the tab to the Royal Silicon app. Use SOQL for live data; no hardcoded record IDs.
Housekeeping:These are throwaway teaching artifacts. After you've seen your workbench work, ask your assistant to "delete the abcXxx LWC and CustomTab from sandbox1" so the org stays clean for the next cohort — or leave it for the facilitator to sweep.
Paste-and-Run · n8n Agent Chat
Headless Prompts.
Ten ready-to-run prompts you can paste straight into the n8n agent chat. Each one drives the Salesforce MCP tools to read live content from the org — no clicking through Salesforce — and returns an answer grounded only in what the records actually say.
How this is different from the Coding tab
The Coding Prompts tab uses an assistant that writes and deploys new components. This tab is pure read-and-reason: the n8n agent reaches Salesforce through the native MCP Server tools (discover, describe, dispatch_readonly) wired via the MCP OAuth2 credential you set up on the Admin Setup tab. You are not building anything — you are asking the org questions in plain English and letting the agent fetch the data and answer.
Every prompt below ends with a guardrail clause — "if a field is blank, say Data Needed and stop." That single line is what keeps a headless agent from inventing values or looping forever when the org data is sparse. Keep it when you adapt a prompt.
H
Ten Paste-Ready Prompts
Paste one · ~1 min each
Copy a prompt, paste it into the n8n agent chat, and swap <OPPORTUNITY NAME> for one of the ten canonical seat deals (Dell PowerEdge XE9680, HPE ProLiant DL384, Supermicro SYS-821GE-TNHR, Lenovo ThinkSystem SR685a, Cisco UCS C885A, and so on). No record IDs required — the tool resolves by name.
Read-and-reason prompts for the n8n agent
1 · Deal Snapshot
Use the Salesforce tool to pull <OPPORTUNITY NAME> with its linked Account and Quote. Give me a one-screen snapshot: stage, amount, close date, quote grand total, the OEM server model, and the single next best action this week. If any of those fields is blank, list it under Data Needed and stop — do not guess or keep searching.
2 · Quote Line Breakdown
For <OPPORTUNITY NAME> and its linked quote, use the Salesforce tool to list every quote line item with product, quantity, list price, net price, and the discount % computed as (List − Net) / List. Add a totals line. If the quote has no line items, say Data Needed and stop.
3 · Discount & Approval Check
Using the Salesforce tool, read the quote lines for <OPPORTUNITY NAME>. Tell me the deepest discount on any line, whether that discount is inside what has already been granted, and the likely approver role. Then write a two-sentence approval note I could send. If threshold or approval fields are missing, say Data Needed and continue with what you have.
4 · Risk & Next Step
Use the Salesforce tool to read the Opportunity <OPPORTUNITY NAME> including Description, NextStep, StageName, and CloseDate. Summarize the top three risks to closing this deal and the one action that most moves it forward. Base every risk ONLY on what the record says — if a risk is not supported by a field, do not list it. If NextStep and Description are both blank, say Data Needed and stop.
5 · Competitive Brief
Build a competitive brief for <OPPORTUNITY NAME>. Use the Salesforce tool to read the Opportunity Description, NextStep, and the Account industry and type. Base competitor points ONLY on what is written in those fields. If no competitor is named anywhere in the record, return three discovery questions instead of naming one.
6 · Close-Date Watch
Using the Salesforce tool, get only these named opportunities: <LIST 2–5 SEAT NAMES>. List any whose CloseDate is on or before <DATE> and not already Closed, showing stage, amount, and whether it looks at risk based on NextStep. Do not pull any opportunities other than the ones I named.
7 · Account 360
Use the Salesforce tool to read the Account behind <OPPORTUNITY NAME>, then list that Account's open opportunities with name, stage, amount, and close date, plus total open pipeline. Note which server models appear across the deals. If the Account has only this one opportunity, say so plainly rather than inferring others.
8 · Data-Quality Audit
Audit data quality for <OPPORTUNITY NAME> using the Salesforce tool. List which of Amount, Type, LeadSource, NextStep, and Description are blank, and give one line each on why that blank matters for a pricing or forecast decision. Do not fill in or guess any of the missing values.
9 · Customer-Ready Email
Use the Salesforce tool to read <OPPORTUNITY NAME>, its Account, and its Quote. Draft a short, professional follow-up email to the customer that references the specific server model and quote total from the record and proposes the NextStep as the agreed action. Keep it under 120 words. If the quote total or NextStep is blank, say Data Needed and stop instead of writing the email.
10 · Strict JSON Output
Return the result for <OPPORTUNITY NAME> as strict JSON only — no prose, no markdown — with keys opportunityName, account, serverModel, stage, amount, quoteGrandTotal, closeDate, topRisk, nextAction. Use the Salesforce tool to fill each value and use null where a field is unavailable. Keep topRisk and nextAction under 12 words each.
Facilitator Note:Have each table paste one prompt, swap in their own seat's OEM/server deal, and read the agent's answer aloud. Then break one on purpose — delete the "if blank, say Data Needed and stop" clause and point it at a sparse record — to show the room why the guardrail is the whole game for a headless agent.
Deep Dive
Headless 360.
Everything you need to know about what Salesforce announced — and why it matters.
The Receipts
Headless Since Day One.
Every door is an API. The new bit — agents are first-class clients now.
2000
SOAP API
Salesforce launched with an API on day one. Every record, every object — programmable from the start.
00
10
2010
REST API
Modern RESTful access. JSON in, JSON out. The integration floodgates opened.
2018
MuleSoft Acquisition
Every system in your enterprise, connected via APIs. The "integration company" era begins.
18
24
2024
Agentforce API
Build, deploy, and call AI agents programmatically. Agents become API clients.
2026
Platform-Hosted MCP
60+ MCP tools, 30+ skills. Hosted by Salesforce. Plugged into the AI tools you already use.
26
26 years of APIs. One brand-new audience: agents.
Architecture
Same Four Layers. Same Products.
Layer
Product
What it does
System of Context
Data 360
All your trusted business data — unified, real-time, ready for agents. Customer profiles, behavioral data, transactions — all harmonized in Data Cloud.
System of Work
Customer 360
Decades of business logic. Sales, Service, Marketing — your operational backbone, now accessible to AI agents via APIs.
System of Agency
Agentforce
Build, deploy, and manage agents at scale. Simple bots to complex multi-step agents with custom planners.
System of Engagement
Slack
Where humans and agents collaborate. Agents surface insights, ask questions, and take action alongside your team.
The Headline
What shipped at TDX.
The HeadlineGATrust Layer Enforced
Platform-Hosted MCP & Coding-Agent Skills.
60+ MCP tools and 30+ pre-built skills, hosted by Salesforce, plug straight into the coding and chat agents your team already uses. No middleware. Same Trust Layer. Auth is just OAuth.
Claude CodeCursorSlackAgentforce Vibes 2.0DevOps Center MCPNatural Language DevOps
Other releases worth knowing
Rich Experiences
Agentforce Experience Layer
Build a card, decision tile, or workflow once. Render it natively wherever people work.
Same component, every surface — no per-channel rebuild
Trust at Scale
Agent Control at Scale
The "shipping is the starting line" tooling. Pre-launch, in-production, multi-vendor.
Pre-launch: Testing Center, Custom Scoring Evals
In-prod: A/B Testing API, Observability
Multi-vendor: Agent Fabric (one control plane)
Game Changer
Your Apex. Your Flows. Now agent tools.
If it's a global @InvocableMethod or an invocable Flow, it's already exposable as an MCP tool. No new framework.
Apex
Got an @InvocableMethod? It's a tool.
global class CalculateRenewalQuote {
@InvocableMethod(
label='calculate_renewal_quote'
description='Build a renewal quote for an Account'
)
global static List<Quote> run(List<Id> accountIds) {
return RenewalEngine.run(accountIds);
}
}
// any global @InvocableMethod is callable by an MCP client
Flow
Got an invocable Flow? Same thing.
// In Flow Builder
Flow: "Escalate_High_Value_Case"
Type: Auto-launched (Invocable)
Inputs: caseId (Text), reason (Text)
Outputs: ownerId (Text), slackChannel (Text)
// any invocable Flow is callable by an MCP client
