Code Reality Labs
Writing from the House
The company-level view: why we build the infrastructure that grounds AI agents in what's real, how the five products fit together, and the principles behind them. Deep technical notes live on each product's own blog.
A Release Is a Checklist, Not a Date
Arbiter shipped its first public pre-release when its checks passed, not when a calendar said so, and followed it with a fix release the next day. The products behind it now start their own releases from what that one taught.
Read post →Count Only What You Can Prove
A number is only as good as what stands behind it. This cycle, TheAuditor, Arbiter and Curator each refused to count something they could not prove: a clean run, a ticket's cost, and progress measured by a flattering score.
Read post →A Look Should Never Change Anything
An agent spends most of its day looking: searching, reading, checking. Across the stack this cycle, three products made the same promise about that: a look changes nothing, an action cannot pass for a look, and a tool that cannot help never stands between an agent and a read.
Read post →A Cycle Spent Hunting Quiet Failures
Every layer of the stack spent this cycle on release preparation, and every pass went looking for the same thing: results that look complete and are not. Here is what each layer hardened, and what ships next.
Read post →A Cycle Spent on Truth and Proof
Not every layer of the stack moves at once, and that is the point. This cycle the work went to the two everything else stands on: the code truth TheAuditor produces, and the proof BenchProctor holds it to.
Read post →A Benchmark Should Outrun the Tools It Grades
We build a scanner and the benchmark that grades it, and we say so. The benchmark is built to cover more than any single tool, so a score reflects real capability, not a test shaped to flatter.
Read post →An Agent Believes Whatever Comes Back
A person reading a suspicious result squints at it. An agent builds on it. So every tool in the stack is held to one rule: when it could not do something, the output says so.
Read post →Your Code Never Has to Leave the Building
Nearly every mainstream AI coding tool starts by uploading your source to someone else's computer. Local-first is the opposite bet, and in an era of cloud AI it is the durable one.
Read post →You Cannot Grade Your Own Homework
Most AI coding setups are a closed room where the model does the work and then declares it correct. Closing the loop takes three separate moves: know the code, act on it, and prove the result against something you did not grade yourself.
Read post →The Sovereign AI Engineering Stack
Four bottlenecks are throttling AI that writes code at scale. Kill all four at once, on hardware you own, controllable from your phone.
Read post →Better Together, Now in Code
The cross-product wiring shipped. See what changes when your tools quietly help each other, and why none of it ever becomes a dependency you can't escape.
Read post →I Kept Removing the Reasons Agents Are Stupid
The founder story behind five products: I wanted AI context that was correct, compact, persistent, and actionable, and each fix exposed the next thing missing.
Read post →One Task, Five Tools, One Trace
Follow one ordinary bug fix through all five layers of the stack and watch each failure mode you usually live with quietly disappear.
Read post →Introducing Code Reality Labs
Why we build the infrastructure that grounds AI agents in what's real, and why it took five products under one roof.
Read post →Why one company, five products
Each layer we fixed exposed the next missing one. The case for building the AI context stack under a single roof, without making any product a hostage to the others.
Read post →Better alone, unfair together
Every Code Reality Labs product earns its place on its own. Here's what changes when you stack them.
Read post →