Monthly-Retro on Start AI Tools - Presented by Intent Solutions

March 2026: 35 Posts, the Eval Framework That Shipped 10 Epics in a Day, and a Meta-Milestone

Tue, 31 Mar 2026 18:00:00 -0500

March was the month the machinery became self-aware. Not in a dramatic AI sense — in the sense that the content system started tracking itself. The tier classification system was designed, built, and deployed. The blog-backfill pipeline went operational. And the j-rig binary eval framework proved that a well-structured evaluation system can move at sprint speed: ten epics from planning to production in a single day.

Thirty-five posts. Three hundred forty commits. The commit count dropped again — January had 627, February had 446, March has 340 — and that trend is now clearly a signal, not noise. The work is getting more focused. Fewer repositories are active. Each commit carries more weight. And the blog is capturing more of the work, which means the commits-per-post ratio continues to compress.

February 2026: 100% Tier 1 and the Security Audit That Changed Everything

Sat, 28 Feb 2026 18:00:00 -0600

February was an infrastructure month. No dramatic debugging sagas. No novel architectural breakthroughs that demanded long-form treatment. Just relentless execution across security auditing, event indexing, agent identity, and planner self-correction. The result: every classified post landed at Tier 1. That is not a failure of the classification system — it is an accurate reflection of what the work looked like.

When the work is “run the security audit, fix the findings, deploy the patch, verify the fix,” the honest classification is Field Note. Dressing it up as a Deep-Dive would be tier inflation, and the whole point of the system is to resist that temptation.

January 2026: 32 Posts, 627 Commits, and the Month Everything Shipped

Sat, 31 Jan 2026 18:00:00 -0600

January was a demolition month. Thirty-two posts. Six hundred twenty-seven commits. Eight distinct projects with meaningful forward progress, three of them going from zero to shipped. December had higher raw commit volume (1,252), but December was mostly foundation-laying. January was the month things started arriving at production.

The difference between “building” and “shipping” is felt in the texture of the commits. December commits looked like init: scaffold project structure, chore: add CI config, feat: stub out module. January commits looked like feat: deploy SDK to npm, fix: production indexer race condition, release: v1.1.0. The work shifted from potential to delivered.

December 2025: The Highest Gear — 1252 Commits and a Roblox Game

Wed, 31 Dec 2025 18:00:00 -0600

This month broke everything. 1252 commits — more than October and November combined, then doubled. The post count landed between the two previous months, but the quality distribution hit the highest T1 rate yet. And somehow there was time to build a Roblox game and ship a Go CLI on New Year’s Eve.

December answered November’s calibration question: yes, you can increase volume and keep quality high. You just have to be relentless about what earns a post.

November 2025: The Depth Month — Fewer Posts, Triple the Commits

Sun, 30 Nov 2025 18:00:00 -0600

October said “write everything.” November said “ship everything.” The numbers tell the story: post count dropped 60% while commits more than doubled. That’s what depth over breadth looks like in practice.

Velocity Dashboard

Metric	November 2025	October 2025	Delta
Posts Published	18	45	-60%
Commits	516	220	+135%
Posts Classified	16	11	+45%
T1 (Deep-Dive / Case Study)	11 (69%)	5 (45%)	+24pp
T2 (Field Note)	5 (31%)	6 (55%)	-24pp
Avg Confidence Score	0.88	0.89	-0.01
Classification Rate	89%	24%	+65pp

Every trend moved in the right direction. 89% classification rate versus October’s 24% — that was the number one goal and it’s basically solved. The T1 rate jumping to 69% means the posts that did get written were substantial. No more “today I updated a dependency” filler.

October 2025: Month One — Baseline Established

Fri, 31 Oct 2025 18:00:00 -0500

First month of tracking. No previous baseline. No trend lines. Just raw numbers from a month where I launched a marketplace, shipped a multi-agent SDR system, survived a production auth crisis, and still found time to redesign a landing page.

This is the month everything starts getting measured.

Velocity Dashboard

Metric	October 2025	Delta
Posts Published	45	— (baseline)
Commits	220	— (baseline)
Posts Classified	11	—
T1 (Deep-Dive / Case Study)	5 (45%)	—
T2 (Field Note)	6 (55%)	—
Avg Confidence Score	0.89	—
Classification Rate	24%	—
Commits per Post	4.9	—

45 posts in one month is aggressive. At the time I thought “just document everything.” In hindsight, some of those were thin. The tier system hadn’t fully kicked in yet — only 11 of 45 posts got classified. That’s a 24% classification rate, which is terrible. The system was new. I was still calibrating what counts as a real post versus noise.