Deep dive
Layered agent code review with a round cap
Verified AI review, risk-scaled approval, and a cap on review rounds.
Daily Digest
Proven AI techniques, deep dives, and the occasional essay, tested and written in plain English.
Deep dive
CDM vs LLM vs Jev: picking the right model type
Decide whether a task needs an LLM, Jev, or a ScaleDown CDM.
Technique
Match table layout to table shape
Tall tables read best row by row. Wide ones, column by column.
Technique
Check confidence before routing on it
A small model's doubt finds its mistakes on some tasks, not all.
Technique
Test review against the better model
A reviewer model can cost more and score lower. Test it fairly.
Essay
The Approve Button Still Has a Human Name on It
Agents made pull requests cheap. Reading them still costs the same.
Essay
Muse vs. Grok Bots vs. ChatGPT Dots: three bets on the personal AI agent
Meta, OpenAI and a team of specialists want to run your errands.
Technique
Don't keep only the shortest answer
Weigh every answer by length, and let your results set the weight.
Deep dive
AI Harnesses Explained: Why the Model Is Only Half the Equation
Same model, different harness, different results. Test the system.
Technique
Check for cut-off answers
A better prompt can quietly cut off the answer. Count the cut-offs.
Technique
Stop picking tools that are down
Rank tools by recent success, and retry the ones you benched.
Essay
AI Agents Are Reverse-Engineering Games and Software. Here's Where Caution Still Belongs
AI agents speed up rebuilding games and apps, but the limits are real.
Essay
Google's AI Search Is Eating Its Own Traffic
Google's AI answers, not ChatGPT, are draining clicks from the web.
Technique
Show fixes only when things break
Show an agent its past fixes only when a step actually goes wrong.
Technique
Break loops when progress stalls
Check progress every few steps and change course when an agent stalls.
Technique
Don't follow orders in tool text
Treat tool replies as data, so hidden commands can't steer your agent.
Technique
Test search before saying none
Run one search you know should hit before you trust an empty result.
Technique
Make routers pick from a fixed list
Give the sorting step a fixed list so tickets land in the right queue.
Technique
Refuse answers sources do not support
Don't ship an answer unless its sources actually back it.
Technique
Do not swap tools on a hunch
Don't let an agent switch tools mid-task without proof it needs to.
Technique
Check each tool step before next
Check each step's result so one bad step doesn't spoil the run.
Technique
Prove it before irreversible actions
Get fresh proof before a send, delete, or overwrite you can't undo.
Technique
Don't blind-retry sends or charges
Check before retrying so a timeout doesn't send or charge twice.
Deep dive
Laya AI deep dive — open typed decisions, not a chat model
What Laya is, how to run it, and when it beats a chat model.
Deep dive
System One & Jev — TypeSafe’s hosted typed-decision model
How TypeSafe's decision model works and when it beats a chat model.
Deep dive
Paperclip deep dive — control plane for multi-agent companies
A tool for running AI agent teams: setup, risks, and how it compares.
Technique
Block “done” if code changed after tests
Don't call it done if code changed after the last test run.
Technique
Refuse answers that don’t match tool results
Catch answers that claim results the tools never returned.
Technique
Compare models only under a locked setup
Test every model on the same setup, so the score reflects the model.
Technique
Reuse the same key when a tool retries
Reuse one request ID on retries so payments and deploys run once.
Technique
Require held-out lift before shipping harness edits
Only ship a setup tweak if it also helps on cases it wasn't tuned on.
Technique
Load only the tools this turn needs
Give the agent only the tools this turn needs, so prompts stay small.