Curriculum

Next live batches start 4 Jan 2027

The world’s most affordable and extensive GTM engineering school.

60 days to become a GTM engineer: 30 days of live training, then two 15-day build projects. Built for non-technical learners, with no code anywhere. US$200, or US$50 self-paced.

The system you’ll build, live

Passed every check 126Caught before a prospect 48
  • L2Hiring post is 90 days old, too stale to use
  • L6Grader FAIL: email over the 80-word limit

Illustrative simulation. In the program you build these checks yourself, layer by layer.

60 days30 live + 30 building

A 1-hour live session every weekday, then two portfolio projects.

US$200one price

For the full 60 days.

0 codeprompt it, don’t code it

No Python, SQL, JavaScript, terminal or GitHub.

Upcoming batches

New live batches start every two months in 2027, with live, instructor-led classes five days a week. All times are US Eastern.

  • Live on Google MeetEvery class is live on Google Meet. Got a doubt? Unmute and ask the instructor then and there.
  • Every class recordedCan’t make it live one day? Watch the recording whenever you have time.
  • Recordings and notes after every classAfter each class we share the recording and the class notes with you.

All upcoming 2027 batches

New batches start every two months. Times and dates are US Eastern. 20 seats per batch.

StartsBatchUS EasternClass daysInquire
MondayBatch 1January 2027US: 8:30–9:30 AM ESTDays: Mon–Fri
MondayBatch 2January 2027US: 10:30–11:30 AM ESTDays: Mon–Fri
MondayBatch 1March 2027US: 8:30–9:30 AM ESTthen 9:30–10:30 AM EDT from 15 MarDays: Mon–Fri
MondayBatch 2March 2027US: 10:30–11:30 AM ESTthen 11:30 AM–12:30 PM EDT from 15 MarDays: Mon–Fri
SundayBatch 3March 2027US: 9:30–10:30 PM ESTthen 10:30–11:30 PM EDT from 14 MarDays: Sun–Thu
MondayBatch 1May 2027US: 9:30–10:30 AM EDTDays: Mon–Fri
MondayBatch 2May 2027US: 11:30 AM–12:30 PM EDTDays: Mon–Fri
SundayBatch 3May 2027US: 10:30–11:30 PM EDTDays: Sun–Thu
MondayBatch 1July 2027US: 9:30–10:30 AM EDTDays: Mon–Fri
MondayBatch 2July 2027US: 11:30 AM–12:30 PM EDTDays: Mon–Fri
SundayBatch 3July 2027US: 10:30–11:30 PM EDTDays: Sun–Thu
MondayBatch 1September 2027US: 9:30–10:30 AM EDTDays: Mon–Fri
MondayBatch 2September 2027US: 11:30 AM–12:30 PM EDTDays: Mon–Fri
SundayBatch 3September 2027US: 10:30–11:30 PM EDTDays: Sun–Thu
MondayBatch 1November 2027US: 9:30–10:30 AM EDTthen 8:30–9:30 AM EST from 8 NovDays: Mon–Fri
MondayBatch 2November 2027US: 11:30 AM–12:30 PM EDTthen 10:30–11:30 AM EST from 8 NovDays: Mon–Fri
SundayBatch 3November 2027US: 10:30–11:30 PM EDTthen 9:30–10:30 PM EST from 7 NovDays: Sun–Thu

Self-paced · Start any time

Can’t make a live batch? Learn self-paced.

For anyone who can’t find a live batch that fits, or wants to start right away and learn at their own pace. 30 days of access.

  • 30 days of access to all class recordings from the current batches
  • Doubt clearing directly with the instructor
  • Access to the Stable GTM learner community, where learners build real GTM engineering projects

US$50

Want to judge the quality first? Watch sample class recordings or read testimonials

A GTM engineering school, not a video course

Live every weekday, built around one real company, and judged on what you ship.

30 days of live training

A 1-hour live session every weekday, Monday to Friday, plus 2 hours of homework. Five phases, each ending in a portfolio deliverable.

  • Live on Google Meet: unmute and clear doubts in class
  • Every class recorded; recording and notes shared after each one
  • A tested “prompt of the day” for your library
  • About 30 reusable prompts by Day 30

Two 15-day build projects

After training you build two complete systems: a beginner outbound system and an advanced AI-assisted engine with human approval.

  • Project 1: a signal-led outbound system for one segment
  • Project 2: a signal-to-meeting engine with a human in the loop
  • Written pass criteria and a 100-point rubric for each

Prompt it, don’t code it

When a step needs logic, a formula, a workflow or an integration, you describe it in plain English to an AI builder and check the result.

  • Claude, Clay’s Sculptor, Claygent and AI formula generator
  • The AI assistants inside Zapier and n8n
  • The skill: precise instructions and judging the output

One company, one portfolio

You adopt one real B2B company on Day 2 and build everything for it, so your portfolio tells one story with real numbers.

  • Hiring and clients are won by shipped systems, not certificates
  • Five deliverables in training, two projects after
  • Layers over tools: tools change every 6–12 months, layers don’t

Seven layers. Value is lost at the handoffs.

Tools change every 6–12 months; the layers don’t. You learn what each layer is for, then the tool that does it today.

LayerNameIts one job2027 no-code toolsFailure mode to name
L0ContextHold ICP, personas, offers, proof, competitors, voice, banned claimsTools: Clay Knowledge Hub, Claude Projects + Skills, a Google Doc context packAI writes confident, generic or false claims
L1DataFind the right accounts and people, with verified contact dataTools: Clay Search, Audiences, Lookalikes, waterfalls; Apollo; Sales NavigatorPrecise outreach to the wrong companies
L2SignalsDetect the “why now”Tools: Clay Signals, RB2B (US only), job-change and hiring signals, your own CRM dataMessaging about something stale
L3OrchestrationMove data between tools on triggers, with rulesTools: Clay Workflows (built with Sculptor), Zapier or n8n built by AI promptA step silently fails and leads stop moving
L4ActivationReach people: email, LinkedIn, phone, adsTools: Instantly / Smartlead / Clay Sequencer, HeyReach with limits, Clay AdsDomain burned, LinkedIn account restricted
L5System of recordBe the single truthTools: HubSpot (free CRM), Google Sheets as the logDuplicates; routing and reports lie
L6Learning loopMeasure, grade AI output, track cost, run testsTools: AI grader prompts, Sheets dashboards, HubSpot reportsNobody knows what worked or what it cost

Two design rules you’ll repeat all program

  1. Rules first, AI where judgment is needed. If a filter or a formula can do it, don’t use an AI agent.
  2. Every AI output passes a check before it reaches a prospect or the CRM: an AI grader prompt plus a human sample review.

The full curriculum, day by day

Five phases, 30 live 1-hour sessions, five portfolio deliverables. Open any day to see what you learn, build and ship.

Every session0–5 homework review5–20 concepts20–35 live demo35–55 guided lab55–60 exit ticket

Phase 1 · Days 1–6

Mindset, strategy and the context layer

By Day 6 you have a company, a strategy and a context pack that every later AI step reads from.

ShipsStrategy brief + context pack

Day 01GTM engineering in 2027: from tool operator to system designer
Objective
Explain what a GTM engineer owns in 2027 and map 10 live job postings to the layers.
Why it matters
Postings now ask for AI, systems and 6+ tools, not “Clay expert”. Misreading the role means building the wrong portfolio.
Teach
Clay coined the title in 2023; the 2027 market snapshot; GTM engineer vs RevOps vs SDR vs agency operator; the non-coding career paths (outbound/growth GTME, agency, fractional, in-house operator); what “supervising AI” means.
Live demo
Ask Clay’s global agent to build a small play from a one-line goal, then ask Claude to critique the result against the layer model.
Lab
Tag 10 US/EU postings by layer, tools named, and whether code is required.
Homework
Posting-analysis sheet + personal skill-gap list (top 5 gaps).
Mistakes to avoid
Treating Clay as the job; applying to SQL-heavy roles without saying how you’d cover that gap.
Prompt of the day
Here are 10 job descriptions. For each, list the tools named, map each responsibility to one of these layers [paste layers], and say whether the role can be done without coding. Return a table.
Day 02The layers, the workspace, and adopting your company
Objective
Set up the full stack and adopt the one company used for all 60 days.
Why it matters
One company makes every artifact add up to one story. Domains bought today are warm by Day 14.
Teach
The seven layers and handoffs (what passes, who owns it, how fast); the company-selection test: B2B, public pricing, clear buyer, US/EU market, visible hiring and news.
Live demo
Create Clay, HubSpot, Apollo and Claude accounts; set up a Claude Project called “[Company] GTM”; connect the HubSpot and Clay connectors in Claude; buy 2 secondary domains.
Lab
Company selection workshop: pitch your pick in 60 seconds.
Homework
Company one-pager (product, buyer, problem, price) + a hand-drawn layer sketch for this company.
Mistakes to avoid
B2C or enterprise-only companies with no observable signals; skipping the domain purchase.
Prompt of the day
Pressure-test this company as a GTM practice target. Score 1–5 on: clear buyer, public pricing, observable signals, reachable decision-makers, US/EU market. Tell me the weakest point and whether to switch.
Day 03ICP from evidence and TAM by account count
Objective
Derive an ICP from customer evidence and size the market as a count of accounts.
Why it matters
Every AI agent and every filter downstream plans from the ICP.
Teach
Mining case studies, G2 reviews and customer logos; firmographic + technographic + situational criteria; explicit disqualifiers; TAM/SAM/SOM as counts, not dollars; A/B/C tiers.
Live demo
Clay’s natural-language company search; Lookalikes built from 10 customer logos; save the result as an Audience segment.
Lab
Write the ICP sentence and 3 segments; read the TAM count for each.
Homework
ICP doc with 3 segments, disqualifiers and a TAM count per segment.
Mistakes to avoid
Persona = job title only; no disqualifiers; a TAM so big it can’t be personalised.
Prompt of the day
Here are 10 case studies from [company]. Find the patterns in company size, industry, tech stack, trigger event and buyer title. Propose an ICP, 3 segments and 5 disqualifiers. Quote the evidence for each.
Day 04Signals: the “why now”
Objective
Write a signal thesis: the 5 events that predict buying for your company, and why.
Why it matters
Signal-triggered campaigns produce the best reply rates. Signals everyone rents are crowded; your own first-party signals are an advantage.
Teach
Signal types (fit, intent, timing, relationship); first-party vs third-party; how fast signals go stale; the champion-departure play; precision vs volume.
Live demo
Clay Signals on an Audience: job change, new hire, funding.
Lab
Rank 10 candidate signals by expected lift and ease of detection.
Homework
Signal thesis table: signal → source → expected lift → message angle → how many days it stays fresh.
Mistakes to avoid
“Raised funding” as the only signal; no hypothesis to test.
Prompt of the day
For a company selling [product] to [persona], list 10 buying signals. For each: why it creates urgency, where it can be detected without code, how many days it stays relevant, and a one-line message angle.
Day 05Offer, message-market fit and the context pack
Objective
Turn strategy into a context pack that AI tools read before writing anything.
Why it matters
The model is rarely the problem; missing context is. Clay reported its Account Agent used 6.4× fewer tokens and answered 2.6× faster because of good data and context.
Teach
Offer design (problem, outcome, proof, low-friction CTA); positioning; the context pack as 7 short docs: ICP, personas, offers, proof points, competitors, voice, banned claims.
Live demo
Load the pack into Clay Knowledge Hub and into the Claude Project. Generate the same email with and without context and compare.
Lab
Build context pack v1 in a shared Google Doc.
Homework
Finish the pack; save a Claude Skill called “write-in-our-voice”.
Mistakes to avoid
Pasting the marketing site as “context”; no proof points; no banned claims.
Prompt of the day
Interview me one question at a time to build a context pack for [company]: ICP, personas, offers, proof points, competitors, voice rules and banned claims. When done, write each as a separate section of no more than 200 words.
Day 06Ship: strategy brief + context pack
Objective
Present a 2-page strategy brief and the context pack for peer review.
Session plan
0–40 learner presentations · 40–55 peer critique against the rubric · 55–60 fixes list.
Deliverable
Deliverable 1: strategy brief (ICP, segments, signal thesis, offer, success metrics, risks) + context pack + 3-minute Loom.
Mistakes to avoid
Metrics without a baseline; no stated risks.
Prompt of the day
Act as a skeptical VP of Sales at [company]. Read this strategy brief and give me the 5 hardest questions you’d ask before funding this outbound motion.

Phase 2 · Days 7–13

Data and Clay, built by prompting

By Day 13 you have 300 scored accounts and 600 verified contacts, with a cost report. Every formula, filter and research column is written by describing it to Clay’s AI tools.

ShipsScored account intelligence table

Day 07The 2027 data landscape and provider benchmarking
Objective
Choose data providers by measured coverage, accuracy and cost for your segment and region.
Why it matters
Provider quality varies by segment and geography. Clay’s own tests across 4,700+ contacts found different winners for SMB and enterprise.
Teach
Apollo (AI Assistant, agents, Claude connector), Sales Navigator, Clay Search, email finders (FullEnrich, Prospeo, LeadMagic, Findymail, BetterContact), company and hiring data (PredictLeads, Crustdata, Ocean.io), AI web search (Exa), weak EU and India coverage, catch-all domains.
Live demo
Run the same 50 contacts through 4 providers in Clay; compare found rate, verified rate and cost per verified email.
Lab
Repeat the test on your own segment.
Homework
Provider benchmark table for your ICP.
Mistakes to avoid
Trusting vendor accuracy claims; ignoring catch-all results.
Prompt of the day
Here is a benchmark of 4 providers on 50 contacts [paste]. Calculate coverage, verified rate and cost per verified email for each, recommend a waterfall order, and explain the trade-off in 3 sentences.
Day 08Clay fundamentals and Clay economics
Objective
Build a Clay table that costs what you predicted.
Why it matters
Clay’s hiring guide names cost-consciousness as a core skill. AI and enrichment spend grows fast at scale.
Teach
Data Credits vs Actions; bring-your-own-key = Actions only; a fully enriched record ≈ 6–20 Data Credits; credit rollover; table vs Audience vs Workflow and when to use each; the AI formula generator; “run only if” conditions.
Live demo
Forecast the cost of a 200-row table, run it, compare forecast to actual. Generate 3 formulas by prompting.
Lab
Build your first table from the Day 3 Audience.
Homework
Cost model sheet: cost per record, per 1,000 records, per expected meeting.
Mistakes to avoid
Enriching every row before filtering; not checking Clay’s help centre for the current charging rule on empty results.
Prompt of the day
(Clay formula generator) Return ‘Tier A’ if employee count is 50–500 and country is US, UK or Germany and the tech stack includes HubSpot; ‘Tier B’ if only two of these are true; otherwise ‘Exclude’.
Day 09Waterfalls, verification and conditional runs
Objective
Build an email waterfall with ≥70% verified coverage and projected bounce <2%.
Why it matters
These are the pass criteria interview take-homes use for enrichment.
Teach
Ordering providers by cost and segment fit; verification and catch-all handling; run-only-if; deduplication; reusing results instead of paying twice.
Live demo
3-provider waterfall with a verification gate; rerun with a different order and compare cost.
Lab
Waterfall on 300 contacts.
Homework
Coverage and cost report.
Mistakes to avoid
Verifying after sending; buying phone numbers you won’t call.
Prompt of the day
I’ll paste my Clay waterfall settings and results. Find where I’m wasting credits, suggest a cheaper order and the ‘run only if’ conditions to add, in plain English I can copy into Clay.
Day 10Claygent and structured AI research
Objective
Build a research agent that returns structured, checkable answers with sources.
Why it matters
AI research replaced manual prospect research; one Clay GTM engineer cut account research from 85 minutes to 5.
Teach
Claygent Builder (versions, free test runs), auto-mode, fixed answer formats (“Yes / No / Unknown” plus a source URL), Skills inside Claygent; why free-text answers can’t be scored.
Live demo
Build a “pricing model + ICP-fit evidence” Claygent; test on 20 rows; save version 2.
Lab
Build 2 Claygents.
Homework
2 Claygents + a 20-row test log (correct / wrong / unknown).
Mistakes to avoid
No source URL; letting the agent guess.
Prompt of the day
(Claygent) Visit {{domain}}. Answer only from the website. 1) Does the company show public pricing? Yes/No/Unknown. 2) Pricing model: per seat / usage / flat / unknown. 3) Paste the URL where you found it. If not found, answer Unknown. Do not guess.
Day 11Scoring and tiering: fit, intent, engagement
Objective
Build an explainable 3-part score and tier accounts.
Why it matters
Deciding who to contact first is the highest-leverage output of the role.
Teach
Point-based scores built with the formula generator; AI scoring with a written rubric; a “why this score” column; checking the score against 20 known customers and 20 bad fits.
Live demo
Score 300 accounts; check how many known customers land in Tier A.
Lab
Tune weights until ≥15 of 20 customers land in Tier A.
Homework
Scored, tiered table + one-paragraph calibration note.
Mistakes to avoid
Black-box AI scores with no reason; never testing against real customers.
Prompt of the day
Score this account 0–100 for fit using this rubric [paste]. Return: score, the 3 strongest reasons, the biggest risk, and one sentence a sales rep can read.
Day 12Connecting tools without code
Objective
Move data from Clay into HubSpot and Google Sheets using native integrations, and build one simple automation by prompting.
Why it matters
Most leaks happen between tools. You must make handoffs reliable without writing integrations.
Teach
Clay’s native HubSpot and Google Sheets actions; field mapping and matching by domain or email; Zapier/n8n triggers and actions in plain words; asking an AI builder to create a flow from a description; testing with 5 rows first.
Live demo
Clay → HubSpot company upsert by domain; then describe “when a row turns Tier A, add it to a Google Sheet and post in Slack” to an AI builder and test it.
Lab
Build the same two handoffs.
Homework
A one-page handoff map: what moves, from where, to where, matched on what.
Mistakes to avoid
Matching companies by name instead of domain; testing on the full list.
Prompt of the day
Build me an automation: when [trigger] in [tool], do [action] in [tool]. Match records on [field]. If the record exists, update it; never create a duplicate. Tell me exactly which steps to click and what to test.
Day 13Ship: scored account intelligence table
Objective
Finish, QA and peer-review your account intelligence table.
Session plan
0–15 QA checklist · 15–45 finish the build · 45–60 peer review.
Deliverable
Deliverable 2: 300 scored accounts + 600 verified contacts, provider benchmark, cost per record, coverage, scoring explanation, 4-minute Loom.
Pass criteria
≥70% verified email coverage; projected bounce <2%; actual cost within 20% of forecast.
Prompt of the day
Audit this table export [paste sample]. Find duplicates, missing fields, suspicious emails and rows that break my ICP rules. Return a fix list sorted by impact.

Phase 3 · Days 14–19

Outreach that survives 2027 inbox and platform rules

By Day 19 you have warm sending infrastructure, three signal-based sequences that passed an AI quality check, and a multichannel plan with risk limits.

ShipsDeliverability runbook + launch-ready campaign

Day 14Deliverability under enforcement
Objective
Design sending infrastructure that meets Gmail, Yahoo and Microsoft rules.
Why it matters
Providers now reject non-compliant mail instead of just filtering it. Gmail stepped up enforcement in November 2025; Microsoft’s Outlook.com rules apply since May 2025.
Teach
Secondary domains; mailbox math (monthly volume ÷ ~25–30 sends per inbox per day); SPF, DKIM and DMARC as “ID cards” for your domain; one-click unsubscribe; spam complaints under 0.3% (target <0.1%); bounces under 2%; Google Postmaster Tools.
Live demo
Set up DNS records using the email provider’s guided wizard; run an inbox-placement test.
Lab
Mailbox math for 1,500 contacts a month.
Homework
Infrastructure plan (domains, mailboxes, daily volume, warmup dates).
Mistakes to avoid
Sending from the main company domain; skipping DMARC; launching before warmup ends.
Prompt of the day
I want to contact [N] people a month over [X] working days. Calculate domains and mailboxes needed at 25 sends per mailbox per day, and give me a warmup and ramp schedule week by week.
Day 15Sequencers and sending discipline
Objective
Configure a sequencer and pick between Instantly, Smartlead, lemlist and Clay Sequencer.
Why it matters
Discipline matters more than the tool. Consistent volume beats bursts.
Teach
Warmup and ramp; inbox rotation; sending windows by prospect time zone (key for India-based senders); reply detection; open tracking off (Hunter’s 2026 data on 31M emails: 7.4% reply rate without tracking vs 4.4% with); Clay Sequencer keeps data fresh at send time.
Live demo
Build a 4-step sequence and push contacts from Clay.
Lab
Configure your sequencer; confirm warmup is running.
Homework
Screenshot of settings + a one-page sending policy.
Mistakes to avoid
Open tracking pixels on cold sends; links and images in the first email.
Prompt of the day
Review these sequencer settings [paste or screenshot]. Flag anything that risks deliverability: volume, timing, tracking, links, signatures, unsubscribe. Rank by risk.
Day 16Copy that works in 2027: relevance over personalisation
Objective
Write sequences built on signals and offers, not first-name tricks.
Why it matters
Instantly’s 2026 benchmark puts the average reply rate at 3.43% (top 10%: 10.7%+), and 58% of replies come from the first email.
Teach
First email under 80 words; signal → what it means for them → offer → low-friction question; 4–7 steps; one idea per email; the “relevance over personalisation” lesson.
Live demo
Rewrite a generic email into 3 signal-specific versions.
Lab
One sequence per signal from the Day 4 thesis.
Homework
3 sequences in the portfolio doc.
Mistakes to avoid
Opening with compliments; asking for 30 minutes in email one.
Prompt of the day
Using our context pack, write a 4-step sequence for [persona] triggered by [signal]. Email 1 under 80 words: observation, implication, offer, one easy question. No flattery, no feature lists, no ‘quick call’. Give 2 subject lines per email.
Day 17AI personalisation at scale, with an AI grader
Objective
Generate emails with AI and check them with a second, separate grading prompt.
Why it matters
Unchecked AI messaging is what burned AI-SDR customers’ domains and reputations.
Teach
Prompt = context pack + row data + fixed output format; the grader rubric (accurate, relevant, no banned claims, under word limit, on voice); a 50-row “golden set” of good examples; pass-rate targets; humans review a sample every time.
Live demo
Claude or a Clay AI column writes 50 emails; a second prompt grades each Pass/Fail with a reason; inspect the failures and fix the writer prompt.
Lab
Iterate to ≥90% pass rate.
Homework
Writer prompt + grader prompt + golden set + before/after pass rate.
Mistakes to avoid
Grading with the same prompt that wrote the email; no golden set.
Prompt of the day
(Grader) You are a strict reviewer. Grade this email against the rubric [paste]. Check every factual claim against the research data [paste]. Return: PASS or FAIL, failed criteria, the exact sentence that failed, and a fix.
Day 18Multichannel: LinkedIn, phone and ads with risk limits
Objective
Add LinkedIn, calls and ads without endangering accounts.
Why it matters
LinkedIn now acts at the vendor level. In March 2026 it removed HeyReach’s company page and restricted its executives’ personal profiles.
Teach
Safe LinkedIn limits (~25/day, ~100/week; one tool per account; never automate a new account); manual-assisted LinkedIn; when phone numbers are worth paying for; Clay Ads audiences.
Live demo
Map a 14-day email + LinkedIn + call cadence; sync a Clay Audience to LinkedIn Ads.
Lab
Map your own cadence.
Homework
Multichannel cadence with a risk note per channel.
Mistakes to avoid
Two automation tools on one LinkedIn profile; pitching in the connection request.
Prompt of the day
Design a 14-day multichannel cadence for [persona] after [signal]: email, LinkedIn and one call. For each touch give day, channel, goal and a one-line message. Keep LinkedIn under 20 actions per day.
Day 19Ship: deliverability runbook + launch-ready campaign
Objective
Ship the infrastructure and campaign you will launch in Project 1.
Deliverable
Deliverable 3: deliverability runbook (DNS, mailbox math, warmup, monitoring, what to do if bounces spike) + campaign (3 sequences with grader report, multichannel cadence) + Loom.
Pass criteria
All domains pass SPF/DKIM/DMARC; grader pass rate ≥90%; written stop rules (pause if bounces >2% or complaints >0.1%).
Prompt of the day
Turn my notes [paste] into a deliverability runbook a new team member could follow: setup checklist, daily checks, weekly checks, and an incident plan with exact thresholds.

Phase 4 · Days 20–25

Signals, no-code automation and the CRM

By Day 25 you have a signal-triggered play that runs on its own, writes to HubSpot and alerts Slack, all built without code.

ShipsSignal-triggered play with CRM write-back

Day 20Signal sources in practice (and the 2026 vendor shake-out)
Objective
Wire three signal sources for your company and measure how often each is real.
Why it matters
Signal vendors consolidated fast in 2026, so systems must survive a vendor disappearing.
Teach
Website visitor ID (RB2B is US-only for person-level data); job changes (Clay Signals, UserGems, Champify); hiring and funding via Clay Signals and Claygent; the 2026 consolidation of signal vendors.
Live demo
Hiring-signal play: Claygent reads a company’s careers page → a fixed answer (“hiring SDRs: Yes/No”) → a scored row.
Lab
Set up 3 signal feeds.
Homework
Hand-check 30 signal events: real / false / stale. Report precision.
Mistakes to avoid
Paying enterprise prices for signals you can’t act on; assuming EU visitor ID works like the US (it needs consent).
Prompt of the day
Here are 30 signal events [paste]. For each, decide: real buying signal, false positive, or too old to use. Explain in one line and give me the precision rate.
Day 21Clay Audiences and Workflows, built with Sculptor
Objective
Move a play from a table to an always-on Workflow over a live segment.
Why it matters
Workflows run at Audience scale with no row limits and show a trace for every record.
Teach
Dynamic segments; triggers (joins segment, schedule, webhook from a form); branches; AI steps vs rule steps; run traces; take a snapshot before editing; describing the Workflow to Sculptor in plain English; test on 10 records.
Live demo
Rebuild Day 11’s scoring table as a Workflow by prompting Sculptor.
Lab
Build one Workflow and read its trace.
Homework
A Workflow + trace screenshots + a one-paragraph explanation.
Mistakes to avoid
Adding AI steps where a rule would do; no small test run.
Note
Workflows and the global agent were in beta in November 2026; status is checked before each cohort.
Prompt of the day
(Sculptor) Build a workflow: when an account joins the ‘Tier A – Hiring SDRs’ segment, find the VP Sales and Head of RevOps, verify emails with the waterfall, write a first email using the Knowledge Hub, and stop if no verified email is found.
Day 22Automation beyond Clay: Zapier or n8n by prompting
Objective
Build an automation Clay alone can’t do cheaply, by describing it to an AI builder.
Why it matters
n8n appears in 41% and Zapier in 40% of postings (Onfire). Most clients have tools outside Clay.
Teach
Triggers, actions, filters and paths in plain words; n8n’s AI assistant and AI Agent node; Zapier’s AI builder; human approval steps; a Google Sheet as the event log; the “no duplicates” rule: always search before create.
Live demo
HubSpot form submission → look up in Sheet → if new, add row → Slack message with Approve/Reject buttons.
Lab
Build the same flow and run it twice with the same lead to prove no duplicate is created.
Homework
Automation + duplicate test screenshot.
Mistakes to avoid
Letting an AI agent do routing that a simple rule can do; no duplicate check.
Prompt of the day
I don’t code. Walk me step by step through building this in [n8n/Zapier]: [describe flow]. For each step tell me the node or app, what to fill in each field, and how to test it. Add a step that checks for an existing record before creating one.
Day 23CRM as the single source of truth
Objective
Design the HubSpot data model and write enriched data back safely.
Why it matters
AI automates a messy CRM’s mess faster. Clean structure comes first.
Teach
Companies, contacts, deals and how they link; custom properties; lifecycle stages and lead status; property ownership (never overwrite what reps typed); matching by domain and email; HubSpot’s connector in Claude; Breeze agents; Salesforce and Agentforce in one slide.
Live demo
In Claude, ask the HubSpot connector to create 5 custom properties and update 50 companies from the Clay export.
Lab
Write a property dictionary.
Homework
CRM system map + property dictionary + successful write-back of 50 records.
Mistakes to avoid
Duplicate companies created by name; automation overwriting rep-entered fields.
Prompt of the day
Design a HubSpot property dictionary for an outbound program at [company]: property name, type, allowed values, who/what writes it, and whether automation may overwrite it. Keep it under 25 properties.
Day 24Inbound: speed-to-lead routing
Objective
Build an inbound router that enriches, scores, assigns and alerts in under 5 minutes.
Why it matters
Clay’s Claygents filled demo forms at 6,346 companies and most never replied. Fast follow-up is an open opportunity and an easy client win.
Teach
Form → enrich → score → route (territory, round-robin, existing owner) → Slack alert → sequence; response-time targets; a fallback owner.
Live demo
HubSpot form → Clay Workflow → owner assigned → Slack.
Lab
Test with 20 fake leads; log the time for each.
Homework
Router + timing log.
Mistakes to avoid
No fallback owner; routing on fields that aren’t enriched yet.
Prompt of the day
Write routing rules for inbound leads at [company] in plain English: by region, company size and existing owner, with a fallback. Then turn them into a decision table I can paste into a Sheet.
Day 25Ship: signal-triggered play with CRM write-back
Objective
Ship a play that runs on a schedule, end to end, with a human approval step.
Deliverable
Deliverable 4: one signal → enrichment → score → message → human approval in Slack → HubSpot write-back, running on a schedule, with trace screenshots, a duplicate test and a Loom.
Pass criteria
Signal to Slack alert under 1 hour; zero duplicates when the same signal is replayed; every CRM write logged in the Sheet.
Prompt of the day
Here is my play’s architecture [paste]. List every handoff, what could fail at each one, how I’d notice, and the fix. Return a table.

Phase 5 · Days 26–30

AI agents by prompting, guardrails and the learning loop

You build AI agents and rep-facing tools entirely through prompts, Claude Projects, Skills and connectors, then put them into production safely.

ShipsPortfolio page + Project 1 plan

Day 26Prompt engineering for GTM: the core skill
Objective
Write prompts that give the same quality answer every time, on 100 rows, not just once.
Why it matters
In a no-code program, prompting is the engineering. A vague prompt at scale is a bug at scale.
Teach
The 6-part GTM prompt: role, context, task, rules, output format, examples; “Unknown” instead of guessing; splitting one big prompt into steps (research → decide → write → grade); versioning prompts in a prompt library; a cheaper model for simple classification and a stronger one for writing.
Live demo
Take a weak prompt, run it on 20 rows, then fix it part by part and rerun; compare pass rates.
Lab
Rebuild your 3 most-used prompts in the 6-part format.
Homework
Prompt library v1 (every prompt from Days 1–25, versioned, with pass rates).
Mistakes to avoid
One giant prompt doing five jobs; no examples; no version history.
Prompt of the day
Here is my prompt and 5 outputs where it failed [paste]. Diagnose why each failed, rewrite the prompt using role / context / task / rules / format / examples, and give me 3 test cases to check the fix.
Day 27Claude Projects, Skills and connectors: rep-facing tools without code
Objective
Package a GTM workflow as a reusable Claude Skill that works with live CRM and Clay data.
Why it matters
Anthropic’s own GTM team packaged Skills and connectors into a Sales plugin, and about 80% of its sales org was using it within months.
Teach
Claude Projects (shared instructions + context pack); Skills (a saved, named procedure with instructions and examples); connectors to HubSpot, Clay and Apollo; read-only vs write permissions; Clay’s Skills Marketplace; writing usage instructions so a sales rep can use the tool.
Live demo
Build an “Account brief” Skill: pulls Clay data + HubSpot history + context pack and returns a one-page brief with talk track and 3 discovery questions.
Lab
Build one Skill and have a peer use it cold.
Homework
Published Skill + one-page usage guide + 3 example outputs.
Mistakes to avoid
Skills with no context; giving write access when read-only would do.
Prompt of the day
Create a Claude Skill called ‘account-brief’. When given a company domain, pull the company and contacts from HubSpot, add Clay research, and write a one-page brief using our context pack: why now, likely pains, who to contact, opening line, 3 discovery questions. Never invent data; mark gaps as Unknown.
Day 28AI agents with guardrails: human approval, grading, cost and logs
Objective
Put an AI agent into production safely.
Why it matters
Clay’s hiring guide names making AI output checkable and controlling AI cost as core skills. Clay’s own team ships plays “like software”, with tests before rollout.
Teach
Slack approval loops (approve / edit / reject with a reason); grader gate before human review; a cost ledger per lead in Google Sheets; budget caps; logging every agent decision; staged rollout (10% → 50% → 100%); a weekly review of rejected drafts.
Live demo
Agent drafts → grader scores → Slack approve/edit/reject → approved drafts pushed to the sequencer → everything logged in the Sheet.
Lab
Add approval and logging to your Deliverable 4 play.
Homework
Approval loop live + cost and grader logs for 25 leads.
Mistakes to avoid
Auto-sending without review; no spend cap; no record of what the agent decided.
Prompt of the day
Here are 20 drafts my team rejected, with their reasons [paste]. Group the reasons into patterns, then propose specific changes to my writer prompt and grader rubric to prevent each pattern.
Day 29Measurement, attribution, compliance and pricing your work
Objective
Report results credibly and operate legally across the US, UK and EU.
Why it matters
Clients and hiring managers pay for measured results. 72% of GTM engineers in the State of GTME survey report measurable revenue impact.
Teach
The funnel (delivered → replied → positive reply → meeting → opportunity); cost per meeting; holdout groups; opens are not success. Compliance: CAN-SPAM (US), UK ICO legitimate interests, France’s CNIL rules, EU visitor-ID consent. AEO awareness. Pricing your work: retainers, per-campaign and fractional rates.
Live demo
HubSpot report + Google Sheets dashboard; Claude writes the weekly summary from the numbers.
Lab
Build your dashboard and a compliance checklist.
Homework
Dashboard + compliance checklist + personal rate card.
Mistakes to avoid
Reporting open rates; running US playbooks on EU contacts.
Note
Compliance content is educational, not legal advice.
Prompt of the day
Here are this week’s campaign numbers [paste]. Write a 150-word update for a client: what happened, what it cost per meeting, what we’ll change next week, and one risk. No vanity metrics.
Day 30Demo day, certification and project kickoff
Objective
Demo your work to a panel and kick off Project 1.
Session plan
0–40 learner demos · 40–52 panel feedback · 52–60 Project 1 kickoff and assignments.
Deliverable
Deliverable 5: portfolio page (Notion or Google Sites) with Deliverables 1–4, Looms, the prompt library and the Project 1 plan.
Prompt of the day
Act as a hiring manager for a GTM engineer role. Review my portfolio page [paste]. Tell me what would make you invite me to interview and what would make you skip me.

Then you build two real systems

Days 31–60. Two 15-day projects for your adopted company, each with written pass criteria and a 100-point rubric.

Project 1 · Beginner · Days 31–45

The Hiring-Signal outbound system for one segment

For your adopted company (or a real client who agrees in writing), build and launch one complete outbound system for one segment and one signal: accounts that are actively hiring for the role your product helps. Built in Clay, the sequencer, HubSpot and Claude, with no code.

It forces every fundamental into one chain (strategy → data → signal → copy → deliverability → CRM → results) and ends in real numbers. Hiring signals are free to detect with Claygent on careers pages, so no paid signal tool is needed.

  1. Day 01Scope and infrastructure1-page brief (segment, signal, offer, targets); 2 domains and 4–6 mailboxes warming
  2. Day 02Signal definitionExact rule for “hiring” (which roles, how many, posted within how many days) + success metrics
  3. Day 03Account sourcing500 accounts in a Clay Audience, with source noted
  4. Day 04Hiring-signal detectionClaygent on careers pages: hiring Yes/No, count, role titles, source URL; precision check on 30 rows
  5. Day 05Fit scoringTiered list; top 150 signal-positive accounts; calibration note
  6. Day 06Contacts2 personas per account → ~300 contacts
  7. Day 07WaterfallEmail waterfall + verification; coverage and cost report
  8. Day 08Research2 Claygents per account (e.g. current sales stack, team size) with sources
  9. Day 09Context pack v2Pack updated for this segment: pains, proof, objections
  10. Day 10Copy v12 sequences × 4 steps; first email under 80 words, built on the hiring signal
  11. Day 11AI personalisation + graderWriter prompt + grader prompt; 50-row golden set; ≥90% pass
  12. Day 12Sequencer + CRMSequencer built; HubSpot sync of contacts, sequence status and replies; no duplicates
  13. Day 13Pre-flight and launchDNS checks, inbox-placement test, stop rules; first 100 contacts sent (or a seed-list launch)
  14. Day 14Reply handlingClaude classifies replies (positive / objection / not now / unsubscribe) and HubSpot is updated
  15. Day 15ReportResults memo, cost per record and per reply, 5-minute Loom, case study

Pass criteria (all must pass)

  • ≥70% verified email coverage
  • Bounce rate under 2%
  • Hiring-signal precision ≥80% on 30 hand-checked rows
  • Grader pass rate ≥90%
  • SPF, DKIM and DMARC pass on every sending domain
  • Every contact traceable from source to CRM
  • Cost per record and per reply reported
  • Memo with 3 lessons and the next experiment

Rubric (100 points) Strategy 20 · Data quality 20 · Deliverability 15 · Copy and grader 15 · CRM integrity 10 · Measurement 10 · Communication 10

Stretch goals A/B test two signals (hiring vs recent funding); add a manual-assisted LinkedIn step; compare Clay Sequencer with Instantly/Smartlead on the same list.

Project 2 · Advanced · Days 46–60

The Champion & Signal Engine: a no-code AI SDR with a human in the loop

Build an always-on engine that watches 3 signal sources, researches and scores each account, drafts multichannel outreach with AI, checks every draft with an AI grader, sends it to a human in Slack for approval, then writes everything to HubSpot and a cost ledger and produces a weekly learning report.

It is what generic “AI SDR” products promised and failed to deliver safely. It proves the skills Clay’s hiring guide values most: reliable handoffs, AI output you can check, human control and cost awareness, and mirrors the “signal engine” take-home that 2026 interview guides benchmark.

The engine: 9 steps, 1 human gate, 1 learning loop

  1. A signal fires and lands in a “Signals log” Sheet with a unique key, so a repeated signal is ignored
  2. A Clay Workflow enriches the account and finds 2 contacts with verified emails
  3. A research Claygent answers fixed questions with sources; a rule score + AI fit score decide whether to continue
  4. A writer prompt drafts an email and a LinkedIn note, and picks the channel
  5. A grader prompt checks each draft; failures get one rewrite, then a “needs work” list
  6. Passing drafts go to Slack with Approve / Edit / Reject; reject requires a reason
  7. Approved drafts go to the sequencer; contact, company, signal and status are upserted in HubSpot
  8. Replies are classified by Claude and update HubSpot and the log
  9. Every step writes cost to a “Cost ledger” Sheet; each Friday a Claude Skill writes the learning report
  1. Day 01ArchitectureArchitecture map, the 3 Sheets (Signals log, Drafts & approvals, Cost ledger), HubSpot properties
  2. Day 02Signal source 1Champion moved, feeding the Signals log with unique keys
  3. Day 03Signal sources 2 and 3Hiring and funding/leadership change; precision check on 50 hand-labelled signals
  4. Day 04EnrichmentClay Workflow for enrichment + contacts, cost logged per account
  5. Day 05Research and fitResearch Claygent with fixed answers + sources; fit score with “why” column
  6. Day 06Writer promptChannel choice; signal-specific angles per signal type
  7. Day 07GraderGrader prompt + 50-row golden set + a list of failure types
  8. Day 08Human approvalSlack Approve / Edit / Reject with reason capture
  9. Day 09CRM write-backHubSpot upsert by email/domain; never overwrite rep fields; activity logged
  10. Day 10Send and repliesSequencer push + reply classification
  11. Day 11Replay testFire the same 20 signals again; prove zero duplicate contacts, drafts or emails
  12. Day 12Cost controlsCheaper model for classification, budget caps, cost per approved lead
  13. Day 13Learning loopLearning-report Skill; ship one prompt change to 10% of leads first
  14. Day 14Safety and complianceOpt-out handling, EU/UK rules, read-only vs write permissions, who can approve
  15. Day 15Demo day5-minute Loom, case study, architecture map, logs

Pass criteria (all must pass)

  • Signal to Slack approval request in ≤15 minutes
  • Signal precision ≥80% on 50 labelled examples
  • Zero duplicates on the replay test
  • Grader pass rate ≥90% before human review
  • 100% of sends approved by a human
  • Cost per approved lead reported
  • Complete log of every signal, draft, decision and CRM write

Rubric (100 points) Architecture and handoffs 20 · Reliability 20 · Prompt and grader quality 20 · Human approval and governance 10 · Cost control 10 · Learning loop 10 · Communication 10

Stretch goals Add the inbound path with under-5-minute response; package the research step as a Claude Skill reps use on demand; compare the same play built in Clay Workflows vs n8n.

The no-code stack you’ll learn

Every tool is used through its interface or by prompting. Most run on free plans or trials.

ToolUsed forPlan for learners
ClayData, enrichment, Claygent research, scoring, Audiences, Workflows, SculptorFree plan for practice; a shared cohort workspace for build days
Claude (Pro)Strategy, copy, grading, building prompts, Projects + Skills, connectorsOne seat per learner
HubSpotFree CRM, forms, lifecycle stages, reportsFree
ApolloSecond data source and benchmarkFree
Instantly or SmartleadCold email sending and warmupTrial, then cheapest plan for Project 1
Google WorkspaceSending mailboxes on secondary domains2 domains + 4–6 mailboxes
Zapier or n8n CloudAutomation outside Clay, built by AI promptFree / trial
SlackAlerts and human approvalFree
Google SheetsLogs, cost ledger, dashboardsFree
LoomWalkthrough videos for every deliverableFree
Notion or Google SitesYour portfolioFree

Finished early?

Pick a third portfolio project. Each ends in a number a hiring manager or client cares about.

  • Inbound speed-to-lead routerMedian under 5 minutes
  • Enrichment waterfall benchmark1,000 accounts
  • CRM clean-up and enrichmentDuplicates cut ≥90%
  • Closed-lost revival playAgainst a holdout
  • Rep-facing Claude Skill5 reps, one week
  • AEO visibility monitorWeekly report

Certification you earn with shipped work

Hiring and client work are decided by systems with numbers, not certificates. So ours require both.

Stable GTM Certified

All 5 training deliverables + Project 1 pass.

Stable GTM Advanced

Project 2 pass.

Stable GTM Distinction

Project 2 score of 85 or more + a live result for a real company or client.

A portfolio package per project

A Loom of 5 minutes or less, a one-page case study with real numbers, the architecture map, the prompt library with versions and pass rates, and the logs.

What you’ll earn

Every learner who passes gets a Stable GTM certificate with their name, their tier and a certificate ID they can share on LinkedIn.

Job support for live batch learners

Included with every live batch, not the self-paced plan. We help you get hired; we don’t promise a job.

LinkedIn profile optimisation

We review and sharpen your LinkedIn profile so it presents you as a GTM engineer, with your projects and results up front.

Interview preparation

We prepare you for GTM engineering interviews: how to walk through your portfolio, explain your systems and handle take-home tasks.

A list of companies and agencies that hire GTM engineers

You get our list of companies and agencies that typically hire GTM engineers, so you can reach out to them directly.

Every GTM engineering job, in our community

We run a community where we post every GTM engineering job we find posted online, with direct links to apply, so you can keep applying after the program.

Yogesh Jaiswal, Founder and instructor, Stable GTM

Your instructor

Yogesh Jaiswal

Founder and instructor, Stable GTM

Yogesh teaches every live class himself. He spent 4+ years in sales at companies like Freshworks, Razorpay and Glider AI, so he knows what a sales team needs from a GTM system.

For the past two years he has worked as a GTM engineer with 30+ companies, startups and agencies across the US and Europe, building the kind of systems you build in this program.

in sales
4+ years
as a GTM engineer
2 years
companies, startups and agencies
30+

Sales experience atFreshworksRazorpayGlider AI

Follow @gtmwithyogesh on Instagram

Prerequisite

The only prerequisite: you’re serious about GTM engineering

Every batch gets far more inquiries than seats, so we choose who joins. We work only with people who are passionate about GTM engineering and bring real energy to learning it.

You’re in the right place if

  • You’re passionate about GTM engineering and energetic about learning it.
  • You believe GTM engineering can change your career and help you grow.
  • You’ll show up to the live classes, do the homework and ship both projects.

If that sounds like you, you’re in the right place. We work with serious people, and that’s why our learners deliver good results. No coding experience needed.

Two ways to learn

Join a live batch for the full program, or learn self-paced from the class recordings. Prices in US dollars.

Self-paced

US$50 Price in US dollars

For anyone who can’t find a live batch that fits, or wants to start right away and learn at their own pace. 30 days of access.

Included

  • 30 days of access to all class recordings from the current batches
  • Doubt clearing directly with the instructor
  • Access to the Stable GTM learner community, where learners build real GTM engineering projects

Not included

  • Live classes
  • Job support, including the job posts in the community

Start any time. Non-refundable once you get access.

Roles this prepares you for

Non-technical learners who want to become GTM engineers. The best-fit roles are outbound, growth, campaign and agency GTM engineering, fractional work and in-house “GTM operator” roles. About 41% of postings ask for SQL; we don’t teach it, and we say so upfront. You win these roles with a portfolio that shows strategy, clean data, safe deliverability, tested AI prompts and measured results.

Why GTM engineering, why now

  • 3,300+postings in 2025. Up from 63 in 2024. Early 2026 beat all of 2025. Source: getaob
  • 95%of postings mention AI. AI, LLMs or agents, across 209 postings in August 2026. Source: Onfire
  • 6.4tools per posting. Clay 66%, HubSpot 50%, Claude 45%, n8n 41%, Zapier 40%. Source: Onfire

Questions learners ask us

Something else on your mind? Click Inquire now and our team will get back to you.

People who are passionate about GTM engineering and energetic about learning it. We get many more inquiries than seats, so we choose serious learners who believe GTM engineering can change their career, and who will show up, do the work and ship the projects. You don’t need to know how to code.

Inquire about a batch

Choose a live batch or self-paced, then share your name, email and LinkedIn profile. Our team will get back to you with everything you need to join.

  • January 2027 · Batch 18:30–9:30 AM ESTMon–Fri · Starts 4 Jan 2027 · 20 seats
  • January 2027 · Batch 210:30–11:30 AM ESTMon–Fri · Starts 4 Jan 2027 · 20 seats
  • Later in 2027New batches every two monthsMarch · May · July · September · November
  • Self-pacedStart any timeUS$50 · 30 days of recordings, doubt clearing, community

Live batches: US$200

·

Inquire about Stable GTM

New weekday live batches start every two months in 2027, and self-paced starts any time. Seats go to learners who are serious about GTM engineering.

Phone number

This opens WhatsApp with your details filled in. Tap send in WhatsApp to reach us. We also save your details so our team can reply to you; see our .

Your Stable GTM certificate

Sample shown. Learners who pass earn Stable GTM Certified, Advanced or Distinction.

SAMPLEStable GTMCERTIFICATE ID · SGTM-2027-0001CERTIFICATE OF COMPLETIONThis certifies thatYour Namehas completed the Stable GTM 2027 GTM Engineering Program:30 days of live, instructor-led training and two build projects,and is awardedStable GTM CertifiedYogesh JaiswalYogesh JaiswalFounder and instructor, Stable GTM2027Date of issuestablegtm.com