Synthesis: March 7, 2026
Projects covered
Four projects active today: paulos (heavy), eclectis (medium — autonomous grinds), authexis (light — triage and planning), polymathic-h (housekeeping only).
Key accomplishments
The skopos autonomous pipeline is validated across multiple projects. The pipeline collapse from five stages to three (triage, prep, grind) shipped and was stress-tested with 10+ successful grinds. A single grind agent now carries full context through implement, test, self-review, merge, and close — eliminating the handoff losses that plagued the old pipeline. Median grind cost is ~$1 per issue.
Dependency-aware prep is working. The new blocked label system lets prep recognize when issues depend on other open issues and hold them back. The AgentProvider refactoring chain (#231 → #232 → #233) validated the full flow: prep blocked each issue, the orchestrator auto-unblocked them as dependencies closed, and grinds executed in the correct order. Four integration bugs were found and fixed during validation.
Cross-project testing exposed real bugs. Running skopos on eclectis (not just paulos) immediately surfaced three issues: a lost env var fallback, a 4MB CODEBASE.md (62k lines of build artifacts), and a missing label. These would never have appeared in paulos-only testing. Eclectis also got two autonomous security/quality fixes merged: SQL injection sanitization and proper Sentry error logging.
Authexis v2 is nearly closed. 18 of 20 issues done, due April 1. Only #742 (alternative layouts — design decision) and #797 (grindable) remain. The decision on #742 gates the milestone.
Cross-cutting themes
Testing on your own codebase isn’t enough. The skopos pipeline worked perfectly on paulos but broke immediately on eclectis. Different project structures (nested venvs, frontend build artifacts, different env var conventions) expose assumptions baked into tooling. This is the classic “works on my machine” problem at the infrastructure level.
New pipeline states have high integration cost. Adding the blocked label required changes in four separate places: outcome detection, priority ordering, skip lists, and finalize cleanup. Each was a separate bug discovered through running the loop. Any future label additions will follow the same pattern — the code isn’t structured to make this easy.
Prep quality directly determines grind cost. Well-specced issues grind for $0.50-1.00. Vague specs burn $2+. The prep step is where work is won or lost — the grind agent is execution, not design.
Carry-over
- paulos: #234 should auto-unblock (deps now closed). 7 issues in prep queue. Consider #190 (log DB).
- eclectis: 2 grindable issues remaining (#78, #79). #80 needs prep (admin impersonation design decision).
- authexis: #742 decision gates v2 milestone. 12 grindable issues in queue. Roland’s search terms need input.
- polymathic-h: Edition 13 due Tuesday. Apple Podcasts artwork still stale (day 5).
Risks
- Authexis v2 due April 1 — decision on #742 could expand scope if not handled carefully
- The AgentProvider refactor was done by grind agents modifying their own runtime — tests pass but extended multi-step grinds haven’t been validated on the new code yet
- Edition 13 newsletter has no draft with 3 days to go
Why customer tools are organized wrong
This article reveals a fundamental flaw in how customer support tools are designed—organizing by interaction type instead of by customer—and explains why this fragmentation wastes time and obscures the full picture you need to help users effectively.
Infrastructure shapes thought
The tools you build determine what kinds of thinking become possible. On infrastructure, friction, and building deliberately for thought rather than just throughput.
Server-side dashboard architecture: Why moving data fetching off the browser changes everything
How choosing server-side rendering solved security, CORS, and credential management problems I didn't know I had.
The work of being available now
A book on AI, judgment, and staying human at work.
The practice of work in progress
Practical essays on how work actually gets done.
The second project problem
Your system works. Then you try it somewhere else and it falls apart. The gap between 'works here' and 'works anywhere' is where most automation dies — and most organizations never look.
The smartest code you'll ever delete
The most dangerous kind of waste isn't the thing that doesn't work. It's the thing that works beautifully and shouldn't exist.
The first real user breaks everything
Your product works until someone actually uses it. The gap between 'works in dev' and 'works for a person' is where most systems fail — and most organizations avoid looking.