I have read 30 GTM newsletters every week for two years. Here is what I built on top of them.
Saturday is GTM execution. Sunday is Claude. Both editions scan in five minutes, give you one move, and never ask you to read the originals.
Save this. The two production prompts and the source database visual are the pieces worth your Tuesday morning.
Send it to 1 founder or operator who reads newsletters but has no system for turning the signal into a Monday action.
The 90-second version
I used to open a Chrome bookmarks folder with 30 GTM newsletters every week. Manual. Slow. Nothing compounded.
I built two automated agents on top of them. Saturday is pure GTM execution. Sunday is pure Claude.
Both editions: 4 operator problems, 1 hero move, tiered actions, one steal-this-move play. Scan in 5 minutes.
Below: both this week’s editions, the full source database, and the two production prompts.
Paste either prompt and build your own version this week.
Read time: 9 min
Monday was a public holiday in the Netherlands. Normal rhythm resumes Wednesday.
2 numbers before we go in
The newsletters in my Chrome bookmarks folder. I opened every one manually, every week, for over a year.
The number I read by hand now. The agents do it. I review the output.
The bookmarks folder that started this
Hi, it is Koen Stam and welcome to GTMcraft: The Future GTM Operator. This newsletter is built from 100,000+ GTM signals collected from 100+ operators and founders, combined with 13+ years of my own lessons and failures from the trenches. I write at the intersection of go-to-market practice and AI-powered systems for founders scaling 0 to 10M+ ARR.
100,000+ GTM relevant signals from LinkedIn, Newsletters and Podcasts indexed. 100+ playbooks structured. 3 GTM operator jobs across 3 GTM motions (SMB, MM, ENT). Same recipe. Now opening as a community.
The two agents
Two editions. Two lanes. One format. Never mixed.
Saturday: GTM execution
The agent scans all 30 newsletters every Saturday morning. Lane rule: pure execution only. Pipeline, positioning, retention, org design, outbound, pricing. If a GTM principle survives without an AI tool attached, it stays Saturday. If an insight is inseparable from Claude, it routes to Sunday or gets dropped. Every publish date is verified against a 7-day window before anything gets included.
Output: 4 operator problems, each backed by 2+ verified sources, a hero move, tiered actions, and a 4-week steal-this-move play.
Sunday: Claude
The agent scans 3 dedicated Claude writers every Sunday morning: Ruben Hassid (How to AI), Charlie Hills (MarTech AI), Ruben Dominguez (The AI Corner). Lane rule: Claude-specific only. Workflows, prompts, specs, experiments that actually ran this week. Not AI news. Not feature announcements.
Output: same structure as Saturday. Problems, tiered actions, a 4-week play.
The seam rule is what makes both editions worth reading. A great GTM insight wrapped in a Claude workflow wants to appear in both. The rule says no. Strip the tool, keep the principle on Saturday. If the insight cannot survive without Claude, it runs Sunday only. The discipline of separation is what keeps each edition clean.
The 30+ sources behind Saturday
30+ newsletters. 6 categories. Scanned every week across a 7-day window.
The three steps at the bottom of that visual are the whole system. Scan: read all 30 across the window. Synthesize: group the noise into 3 to 5 real operator problems, each backed by multiple sources. Ship: you get the soundbites, tiered actions, and one steal-this-move playbook.
The categories are the editorial logic. GTM strategy and systems, sales and outbound, marketing and positioning, growth and product, AI and automation, founder and creator. Each cluster has 5 sources. No single cluster dominates a week. The signal has to show up across multiple categories before it becomes a problem block.
This Saturday: lean beats big
11 signals from 30 newsletters. 4 problems. 1 playbook. Here are the 3 that matter most:
Problem 1: your GTM org is carrying weight your 2026 peers already cut Signal: ICONIQ 2026 benchmark across 150+ B2B companies. The median org is running 20-30% leaner, roughly 9x flatter, and producing nearly 2x net new revenue per rep versus two years ago. A $270K productivity gap separates top from bottom performers. Do first: calculate net new revenue per rep. Last 4 quarters of new ARR divided by fully loaded GTM headcount. You know this worked when: you can name your revenue-per-rep number from memory.
Problem 2: your pipeline is fishing in the same 5% as everyone else Signal: Cannonball GTM put a number on it. Only about 5% of your market is actively buying in any quarter. The growth lives in the 15% with acute pain who have not started shopping yet. They bring need without budget or timeline and close at lower rates over 90 days. Do first: score your last 10 booked meetings on pain (1-5) and urgency (1-5). Anything scoring 1-2 on both is not pipeline. You know this worked when: your forecast separates need-4-5 deals and outbound stops getting killed at first review.
Problem 3: your retention number is masking churn that now hits in Year 2 Signal: SaaStr on the shift. The accounts that used to churn in Year 3 are now going in Year 2 because alternatives are easier to adopt. Do first: pull gross logo retention next to net revenue retention. A wide gap means expansion is papering over churn. You know this worked when: you can see a low-engagement account two quarters before renewal, not two weeks.
This Sunday: write the spec before you prompt
8 signals from 3 creators. 3 problems. 1 playbook. Here are the 3 that matter most:
Problem 1: your prompts describe the task. They never say what done looks like. Signal: Ruben Dominguez and Ruben Hassid both traced the same failure. Karpathy’s Autoresearch method works because the spec tells the agent exactly what better means. Without a spec, every output is version 1. The agent only improved because it knew what improvement looked like. Do first: rewrite one prompt you use 3+ times a week. Add one line: “Done looks like: [specific format, length, one success metric].” You know this worked when: a first draft clears your bar with light edits and you stop re-running the same prompt to get what you want.
Problem 2: you run experiments one at a time, by hand. Signal: Charlie Hills on the small version of Karpathy’s loop. Eric Siu took the same loop to marketing and went from 30 experiments to 4,500 per year at a cost of $36,500. You do not need a scoreboard at that scale. You need a spec and a way to compare variants before you ship. Do first: pick one asset you ship weekly. Have Claude produce 5 variants and score each against your spec before you choose. You know this worked when: every shipped asset has a runner-up and a logged reason it lost.
Problem 3: your judgment lives in your head, so nothing runs without you. Signal: the quiet detail in Karpathy’s setup. The human writes exactly one file - the spec - and the system runs the rules from there. Charlie Hills makes the fixes concrete. As long as your criteria live in your head, you grade every draft and nothing compounds while you sleep. Do first: write the 5 rules Claude should apply every time it judges your output. That file is your spec. You know this worked when: a teammate runs the loop and hits your bar without asking you.
The Saturday production prompt
Paste this into a Claude Project or Cowork session, add your source URLs, and it will scan the week, verify dates, identify operator problems, and format them with tiered actions.
You are GTMcraft Signal Editor (Saturday), producing the GTM
Execution edition: highest-signal go-to-market insights from
the past 7 days, organized by operator problems with tiered actions.
Lane rule: pure GTM execution only. Pipeline, outbound, pricing,
positioning, retention, sales leadership, PLG, demand gen,
forecasting, hiring, org design. No AI tactics, no Claude tactics.
Seam rule: if a GTM principle survives without the tool, keep
the principle and strip the tool. If inseparable from a specific
AI tool, route to Sunday or drop it.
Scan window: 7 days ending today, inclusive.
Sources: [paste your newsletter archive URLs]
For each operator problem, output:
- Problem label: specific, "your X" framing
- Synthesis: 3-5 sentences, what the signals say together
- Linked sources: every cited post with URL
- Do first: one action under 30 minutes
- Do this week: 2-3 actions within 5 business days
- Do this month: one systemic change
- You know this worked when: one observable outcome
Rules:
- Each problem requires 2+ verified in-window sources
- Verify every publish date by reading it off the post
- Never trust search snippet dates
- Thin week: 3 problems, not a forced 4. Do not pad.
- No banned words: delve, leverage (verb), robust, holistic,
synergy, unlock, empower, navigate, landscape, foster, harness
- No em dashes, en dashes, exclamation marks
- No "It's not X, it's Y" constructions
- Peer operator voice. Direct. Every sentence earns its place.The Sunday production prompt
Three sources only. Claude-specific only. Same structure, different lane.
You are producing the GTMcraft Signal, Sunday Claude Edition:
the clearest Claude workflow moves for GTM operators this week,
sourced from 3 dedicated Claude writers.
Lane rule: Claude-specific only. Workflows, prompts, specs,
experiments that actually ran this week. Not AI news. Not
feature announcements.
Scan window: 7 days ending today (Sunday), inclusive.
Substack only. Do not scan or cite LinkedIn.
Sources:
1. Ruben Hassid - How to AI - ruben.substack.com
2. Charlie Hills - MarTech AI - charliehills.substack.com
3. Ruben Dominguez - The AI Corner - the-ai-corner.com
Critical: two creators are named Ruben. Always use full names.
Ruben Hassid = How to AI. Ruben Dominguez = The AI Corner.
Verify every publish date by reading it off the fetched post.
Never trust search snippet dates.
For each operator problem, output:
- Problem label: specific, "your X" framing
- Signal: what the source found this week, one sentence
- Do first: one action under 30 minutes
- Do this week: one action within 5 days
- Do this month: one systemic change
- You know this worked when: one observable outcome
Rules:
- Deduplicate across all 3 sources
- Maximum 3 problems. Do not pad a thin week.
- Paywalled posts: use free section only, flag it
- No banned words: delve, leverage (verb), robust, holistic,
synergy, unlock, empower, navigate, landscape, foster, harness
- No em dashes, en dashes, exclamation marks
- No "It's not X, it's Y" constructions
- Peer operator voice. Direct. Every sentence earns its place.Both prompts are the starter versions. The full production build - the 5-file suite with the HTML visual, Circle post, Substack newsletter, PNG render at 820px, and QA scorecard - runs on a more complete instruction set. That full system, the HTML design template, and the QA criteria live inside GTMcraft at gtmcraft.circle.so.
Build the Saturday scanner this week. Pick 5 newsletters from the source database above, paste their archive URLs into the Saturday prompt, and run it on this week’s posts. Reply and tell me what it surfaces that you would have missed reading by hand.
I read every reply.
See you Wednesday,
Koen
Your (human) GTM Agent
