GTMcraft Claude Signal: The Model Stopped Being the Moat
Cheaper to run, faster at your job, suspended in three days. The four things that actually compound now.
Save this. The weekly GTM Claude signal saves you hours of reading and helps you become better with Claude on Monday. It is the one piece worth your Sunday (or Monday) morning.
Send it to 1 founder or GTM operator who is still not mastering Claude for GTM and is struggling implementing any learned AI lessons into action today.
The 90-second version
Your AI bill is mostly re-read, not new work. Most setups resend the same context every call. Prompt caching cuts repeated input on Claude by up to 90%, and a full cleanup drops a typical bill 50 to 90%.
The fix is Monday-sized: lean your context files, start fresh sessions instead of long threads, batch tasks into one message, and route cheap models for cheap work.
The government switched off Anthropic’s most powerful model 3 days after launch. Hassid and Charlie Hills both had to rewrite this week’s guides from Fable 5 to Opus 4.8. If your workflow is wired to one model, that is a single point of failure.
The how-to is now free. Hassid shipped a Skill that walks anyone A to Z, so knowing how stops being the job. Amodei says it plainly: execution commoditizes, judgment survives.
This week’s build: cut your AI bill in 4 weeks. Lean one bloated context file this Sunday.
Read time: ~5 minutes.
Lane note: Saturday is GTM execution, Sunday is Claude. Here is the one move that matters most this week.
This week’s number #1 move
Your AI bill is mostly paying to re-read the same thing. Ruben Dominguez pulled the mechanic this week: you pay per token on the way in and the way out, and most setups resend the same instructions, the same context, and the same files on every single call, paying full freight thousands of times. Prompt caching alone trims that repeated input on Claude by up to 90%, and stacking the rest of the cleanup drops a typical bill by half or more, with output quality held steady. For a lean team watching a Pro or Max plan hit its limit by lunch, this is the cheapest win of the week, because none of it needs a better model. It needs less waste.
Start here:
Open your longest-running Claude session, the one you keep adding to, and start a fresh one for the next task. Every follow-up in a long thread re-reads the whole history and bills you for it again.
The rule:
If a session keeps growing, it keeps re-reading. Start fresh per task and keep the files it reads lean.
Sources:
Ruben Dominguez, Jun 14 (https://www.the-ai-corner.com/p/llm-token-cost-optimization-playbook-2026)
Problem 1: You wired your workflow to one model, and this week one got switched off
Most setups quietly assume the model you use today will be there tomorrow. This week broke that assumption. Anthropic shipped Fable 5, its most powerful model, and the US government ordered it suspended three days after launch on a national-security order. Both creators publishing this week felt it. Ruben Hassid had a finished newsletter on the new model that he could not send. Charlie Hills had to bolt a correction onto the top of his guide, swapping every Fable 5 step to Opus 4.8. If your best workflow only runs on one model, or worse, lives inside one chat’s memory, you have a single point of failure you do not control.
Start here:
List the workflows you would lose if your main model vanished tomorrow. Star the one that would hurt most.
The rule:
If a workflow only exists inside one model’s memory, you do not own it, you rent it. Keep the instructions and context in files you control.
Sources:
Charlie Hills, Jun 14 (https://charliehills.substack.com/p/claude-code-is-terrible-at-design)
Ruben Hassid, Jun 14 (https://ruben.substack.com/p/i-replaced-myself)
Problem 2: The how-to is now free, so knowing how is no longer the job
Ruben Hassid, whose entire newsletter teaches people how to do things with AI, spent this week building a Skill that does his job for him. You type /how-to, describe what you want, and Claude walks you A to Z, even if you have no idea where to start. His point is sharp and a little uncomfortable: the how-to is the easy half, and it is now free. What he still gets paid for is the other half, the mess. The button that will not click. The two days he tested something that did not work so you never have to. Dominguez heard the same line from Amodei this week: as AI takes the execution, the edge moves to judgment and taste. So stop hoarding the how. Package the steps you repeat into a Skill, and spend the time you get back on the part the AI cannot do, deciding what is worth doing and whether the output is any good.
Start here:
Pick one task you explain to people or redo from scratch every week. Write the steps down the way you actually do them.
The rule:
If you can write the steps down, it is a Skill, not your job. Keep the judgment, automate the how.
Sources:
Ruben Hassid, Jun 14 (https://ruben.substack.com/p/i-replaced-myself)
Ruben Dominguez, Jun 18 (https://www.the-ai-corner.com/p/dario-amodei-circuit-documentary-10-takeaways-2026)
Problem 3: Your GTM motion still runs through you at every step
Most lean GTM has one bottleneck, and it is the founder. Ruben Dominguez took apart how Carles Reina, employee number four at ElevenLabs, built a go-to-market toward $500M in revenue, and the useful part is that the playbook is teachable. You can encode an operator’s rules into agents that run the motion: source, enrich, sequence, forecast, expand. The agents inherit the discipline that scaled revenue instead of generic best practice, and your people keep the part that actually closes, the judgment and the room. This lines up with what Amodei told Dominguez the same week: once AI takes execution, the edge moves to go-to-market craft. The mistake to avoid is bolting AI onto a broken funnel. Write down the rules you already run by, hand the repeatable steps to agents, and draw a clear line around the work that stays human.
Start here:
Take one GTM step you run by hand every week, like research, enrichment, or first-touch drafting, and write the rules you follow when you do it well.
The rule:
Automate the steps you can write rules for. Keep the close, the forecast call, and the relationship human.
Sources:
Ruben Dominguez, Jun 20 (https://www.the-ai-corner.com/p/your-gtm-run-by-agents)
Ruben Dominguez, Jun 18 (https://www.the-ai-corner.com/p/dario-amodei-circuit-documentary-10-takeaways-2026)
Also on the radar
Ruben Dominguez, The AI Corner: the SaaSpocalypse is real but misread. $285B of software market cap vanished the day Cowork launched, yet total software grows while distribution shifts. Write the sentence “AI can do everything we do except ___” and make your ending specific. Link
Charlie Hills, MarTech AI: before you ship anything Claude built, ask “what would the taste critic say?” On his own site it caught five ambient animations running at once that read as AI-made. Link
Charlie Hills, MarTech AI: borrow premium design components without inheriting their look. Treat every pasted component as a structural donor: keep the skeleton, swap in your colors, fonts, and words. Link
Ruben Dominguez, The AI Corner: prompt caching is the lever most teams skip. Most apps resend the same system prompt thousands of times a day at full price, when caching cuts that repeated input on Claude by up to 90%. Link
Steal this move: cut your AI bill in 4 weeks
For you if: founder or operator, 0 to 10M+ ARR, watching your Claude plan hit its limit faster each month or your API spend climb, with no engineer to optimize it for you. Four progressive weeks, Measure to Operate. The full checklist is on the card.
Why this matters now
The model is no longer the moat. This week made that concrete from three directions: it got cheaper to run, faster at replacing your execution, and fragile enough for a government to switch off in three days.
So your edge moves to what sits around the model. The waste you cut. The judgment you keep. The files you own that any model can read. The motion you encode so it runs without you.
This builds on Software Got Cheap from June 7. Cheap building raised the question of what to build. This week raises the question of what you keep when the model itself is not yours to count on. It also extends Taste Is the Moat from May 24, with a new line on it: the how-to is now a free Skill, so taste and judgment are what is left to sell.
For a lean team at 0 to 10M+ ARR, the move is the same in every lane. Spend less re-reading, depend less on one model, automate the steps you can write down, and keep the judgment for yourself.
What is your number one takeaway from this week’s GTMcraft Sunday signal?
Reply or send me a DM.
See you inside GTMcraft’s Operator Room,
Koen
Your (human) GTM Agent
