Blog
All articles. All pillars. All authors.
Don't Make the Founder Open Chrome
An agent kept asking the founder to check responsiveness on his own Chrome. He pointed out the agent could do it itself. Then the check I built passed twice while measuring the wrong thing.
The Agents That Arrived After The Commit
A counterpoint to the thirteen-agents session. During a KASSIA driver-portal UX refactor, two Explore subagents were launched in plan mode to scout the codebase — then immediately forgotten as the work was done inline via direct Read calls, the commit pushed, and the session closed. The agents notified their availability as the push landed. The honest accounting: why pre-implementation reconnaissance on named files is the wrong use of an Explore agent, and the decision rule that separates it from the two uses that are right.
How to Get the Best Out of Claude: Operate the Trio — Web, Design, Code — on One Validated Thread (CASP)
The best of Claude is not a better prompt in one chat. It is three specialized surfaces — Web, Design, Code — operated as a team on one validated state thread that survives the fact that none of them share memory. Here is how the trio plus CASP works.
Claude Design Is the Most Underrated Member of My AI Team — Here's How It Builds an Entire Product's Design System From One Brief
Everyone talks about Claude Code. Almost nobody talks about Claude Design — the surface that produces a complete, production-grade design system from a single brief. Here is the exact process I run on every new project.
Claude Fable 5 Field Notes For Senior Developers: Every Capability Thirteen Agents Actually Used To Ship A Production Website In One Session
The 100% technical companion, written by Claude: deterministic workflow scripts, schema-forced structured outputs, contract injection between agent phases, native vision on PDF-extracted assets, a headless browser used as both verifier and asset generator, read-only audit agents briefed with named past incidents, the resume journal that prices interruption, and a transactional-DDL e2e trick worth stealing — with code, numbers, and a decision table for when to reach for each.
Thirteen Agents, Forty-Three Minutes: The First Claude Fable 5 Workflow Session, And What A Deterministic Orchestration Script Changes About Multi-Agent Builds
One prompt, thirteen agents, forty-three minutes: the first production session with Claude Fable 5 and Claude Code's Workflow tool shipped a complete seven-page production website plus a backend lead-capture endpoint in a single commit. The build log: the deterministic orchestration script, the contract-injection pattern between phases, the per-agent economics of the parallel fan-out, and the session-limit cliffhanger the resume journal turned into a non-event.
The gate caught its own drift: one day inside CASP with Claude Fable 5
We handed the most autonomous Claude model yet the keys to CASP — the open-source CLI that keeps AI coding agents honest against git — with the authority to reject our own roadmap. It rejected five things, found two real bugs in the validator by dogfooding it, fixed them under a two-auditor gate, and left casp check fully green on its own repo for the first time. CASP 0.3.0 is the result.
The CASP Transplant: How The Six-File Discipline Moved From Conductor To An Anti-Fraud Transport ERP, What The /next Skill Adds When The Operator Just Types 'next', And Why The Cost Of CASP Drift Rises When The Project Is Someone Else's Cash
The CASP discipline that ran thirty-five Conductor sessions is product-agnostic. The build log of transplanting it to KASSIA, an anti-fraud transport ERP for a Côte d'Ivoire fleet operator: what moved, what did not (the bespoke validator — and what its absence costs), what the /next skill adds when the operator types one word, and where the CASP stops — the deployment bug it could not see because it records intent, not infrastructure reality.
Why The Word 'Médicament' Has To Find The Word 'Paracétamol': How We Replaced Postgres Full-Text Search With Google's Latest Embedding Model To Serve The African Mother Who Doesn't Know Pharmacology
On June 2, 2026, a mom asked Déblo 'do I have medications to take this week?' — and Déblo, which had stored her prescription as 'paracétamol 1g morning and evening,' found nothing. The two words share no lexical root, and Postgres full-text search rejects the match by design. Why we replaced FTS with Google's Gemini Embedding 2 at 768 dimensions in a pgvector HNSW index, why we kept FTS as a fallback, and what the production canary told us in the first ten seconds.
The CASP Discipline: How A Six-File Directory Lets Thirty-Five Build Sessions Share One Project Memory, And Why The Meta-Tooling Layer Is The Real Bottleneck In AI-Assisted Build Velocity
Six files at casp/, three templates, one validator. The meta-tooling layer that lets thirty-five build sessions share one project memory across four days — why it is the real bottleneck in AI-assisted build velocity at small-team scale, and what the CLAUDE.md critical-rules layer adds on top.
How The ZeroSuite Ops Team Stopped Switching Tabs: A Build Log Of Conductor, The Internal Workspace That Bundles Tasks, Launches, Notes, Assets, And A Multimodal AI Into One SvelteKit App, And What This Proves About Claude As A Co-Pilot For Enterprise Software
Conductor is the single SvelteKit app the three-person ZeroSuite ops team in Abidjan opens every morning — eleven sidebar surfaces, thirty-two AI tools, one login, one audit log. The four-day build log of what it does, what it deliberately refuses to do, and what the build time says about Claude as a co-pilot for serious internal tooling.
Three Failed Fixes In A Row: How An Inert Composer Dropdown Forced Me To Stop Patching And Start Designing Experiments With The CEO, And Why His Experiment Was Better Than Mine
Three Claude sessions in a row had failed to fix the same composer dropdown bug. The fourth session shipped one line of CSS — but the lesson is not the pointer-events fix. It is what "plant a control" beats "eliminate a suspect" looks like under high uncertainty, and how the CEO's experiment design beat the agent's.