Brett Stark

AI built. Ship proven. Weekly build logs for experienced operators building real systems with AI here in Chicago.

[Brett Stark @BrettStark]

New post tomorrow!
Missed the internet boom? Don’t miss this one. Get started with AI this week.

https://t.co/tc5MqRwubU
#AISecondAct
#MidCareerReinvention
#BuildWithAI
#ChatGPT
#AIProductivity

[Brett Stark @BrettStark]

Ask AI to diagnose your nervous email, not rewrite it. 'What's the actual message in one sentence? Tone in one word? What would confuse the recipient?' Got back: message is 'no but I feel guilty', tone is 'apologetic'. Cut two paragraphs. Sent.

[Brett Stark @BrettStark]

Before you send the email you're nervous about — paste the draft and ask: 'Read this as the most defensive version of the recipient. Where will they feel attacked?' Usually it's one throwaway line. You delete it. The fight doesn't happen.

[Brett Stark @BrettStark]

Before you send the email you've rewritten four times: ask AI to DIAGNOSE it, not rewrite it. 'What's the actual message in one sentence? What's the tone in one word? What would confuse the recipient?' Four minutes. Still sounds like you.

[Brett Stark @BrettStark]

Before sending the email you've rewritten four times: paste it into Claude with this prompt — 'What's the actual message in one sentence, the tone in one word, what would confuse the recipient. Don't rewrite it.' Four minutes. You fix it. It still sounds like you.

[Claude @claudeai]

Introducing Claude Sonnet 5, our most agentic Sonnet yet.

It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models.

[Anthropic @AnthropicAI]

We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.

We'll begin restoring access tomorrow, and will share an update soon.

We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.

[Brian Armstrong @brian_armstrong]

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching.

Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work.

Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task.

Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented.

Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted.

Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect.

The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable.

Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.

[Brett Stark @BrettStark]

Build a 1:1 Prep Agent in Claude (Custom GPT) | BuildProven https://t.co/7fdXtXKL15

[Brett Stark @BrettStark]

Most AI summaries read identical whether they caught the clause that matters or skipped it.

New BuildProven: a 3-section prompt that makes Claude hand you the summary AND what it left out AND what to re-read yourself.

→ https://t.co/GMYojo1L8S

[Brett Stark @BrettStark]

After a call where 3 people talked past each other, paste the transcript into Claude and ask: 'List decisions made, ones people THINK were made but weren't, and every action item without a clear owner.' That last line catches a six-week delay.

[Brett Stark @BrettStark]

AI summaries lie the same way every time: not by getting facts wrong, but by staying silent about what they cut.

Tomorrow's BuildProven: the 3-section prompt that forces Claude to show you what it left out.

→ https://t.co/WQYMO9nvyd

[Brett Stark @BrettStark]

After a messy call, paste the transcript into Claude/ChatGPT: 'List decisions actually made, ones people THINK were made but weren't, and action items without a clear owner. Flag anywhere two people agreed but meant different things.' That catch stops a six-week delay.

[Brett Stark @BrettStark]

@kunchenguid so, can we use @openclaw with our max sub or not?

[Brett Stark @BrettStark]

Yesterday's: 3 repos. 80 commits in. Same problem, solved 3 different ways.

https://t.co/LFCOxKt9cL…?utm_source=twitter&utm_medium=social&utm_campaign=newsletter&utm_content=nl-three-repos-eighty-0611-fri-tw

[Peter Steinberger 🦞 @steipete]

Here's a simple loop: Tell codex to maintain your repos, wake up every 5 minutes and direct work to threads. That makes it easy to parallelize+steer work as needed.

I use a orchestrator skill combined with my triage+autoreview+computer use skills, so some work can land autonomously. https://t.co/FbBoJTIcfd
https://t.co/8389roVnOm

[Brett Stark @BrettStark]

New BuildProven newsletter out now: 3 Repos. 80 Commits. Same Problem.

https://t.co/LFCOxKt9cL…?utm_source=twitter&utm_medium=social&utm_campaign=newsletter&utm_content=nl-three-repos-eighty-0611-thu-tw

[Brett Stark @BrettStark]

If the AI gives you something wrong, don't argue with it. Tell it the rule it broke, then ask again. You're not debating. You're handing it the constraint it was missing.

[Brett Stark @BrettStark]

3 repos. 80 commits in. Same problem, solved 3 different ways.

https://t.co/LFCOxKt9cL…?utm_source=twitter&utm_medium=social&utm_campaign=newsletter&utm_content=nl-three-repos-eighty-0611-wed-tw

[Brett Stark @BrettStark]

The pattern is now undeniable: three rounds, three times I called something "already there" when it was partial or absent and
each time it was because I described the intent of the code rather than checking what it actually does/contains. Codex's ; it's that it goes and checks.

[Brett Stark @BrettStark]

My invoice-chaser tool runs for about 40 cents a day.

Here's what it does: emails me every client 30 days overdue. Before, I tracked that in my head and forgot half of them.

What would you point it at first?