One week of SDD in production: the numbers
The previous two posts made claims. This one is the data trail, pulled from the GitHub API across five repositories for the week of April 14–21, 2026.
Same workflow in every repo: spec written in Perplexity, implemented by Claude Code, validated by GitHub Actions, merged, deployed.
ericcaskey.com, first week (Apr 14–21, 2026)#
The site launched during this window. Every number below is from week one.
| Metric | Count |
|---|---|
| PRs opened | 12 |
| PRs merged | 11 |
| Production deploys | 6 |
| Total CI pipeline runs | 20 |
| PR validation runs | 12 |
caskeycoding.com platform, cumulative (frontend + backend API + infra)#
| Metric | Count |
|---|---|
| Frontend PRs merged | 27 |
| Frontend production deploys | 18 |
| Frontend CI pipeline runs | 57 |
| Backend API PRs merged | 57 |
| Backend API production deploys | 11 |
| Backend API CI pipeline runs | 100 |
| Infrastructure (CDK) PRs merged | 20 |
| Infrastructure deploys | 6 |
| Infrastructure CI pipeline runs | 45 |
Specs repo, first week (Apr 14–21, 2026)#
| Metric | Count |
|---|---|
| Spec PRs merged | 51 |
Specs repo, cumulative#
| Metric | Count |
|---|---|
| Spec files (markdown) | 138 |
| Total spec PRs merged | 59 |
The number worth pausing on#
51 spec PRs merged in one week. That is not documentation overhead, that is the scaffolding that made the other hundred-plus code PRs safe to merge. 138 spec files drive production code across three code repos and a launch-week site.
Every one of those spec PRs was reviewed, gated by a voice scan and an NDA scan, and merged as a deliberate decision. None of them was generated as a byproduct of code. The spec PR always precedes the code PR, that is the whole discipline.
What to link#
caskeycoding-specs-demo, the curated public demo repo. Point one spec in it: "here's what one of the 138 looks like."- ADR-003, Spec-Driven Development
- ADR-004, SDD File Structure
- Two commercial projects, a financial-position reviewer and an AI marathon coach, are in private development on the same workflow. Brand signal, not demonstration material.
Closer#
The demo is public. The specs behind the numbers are private, on purpose. The rigor is visible either way.
Previous in the series:
Keep reading
Watch the agent write
A polish agent drafts an essay against a pre-approved topic.
Ten days of June: the SDD velocity numbers, seven weeks in
In April I published one week of SDD production numbers. The same data trail rerun for June 1 through 10 shows the velocity curve: 309 PRs opened, 293 merged, about 185 production deploys, and one footnote about outrunning GitHub Actions' default limits.
Autonomy is mostly knowing when to stop
I handed a backlog to Claude Fable, told it once it could merge, and let it run. It shipped seventeen items across five repos. The line that mattered was not in the work it finished. It was in the work it refused to touch.
An orchestration mode is only as good as its backlog
Anthropic published a guide on building a session-level orchestration mode. I built it two ways, on the CLI and on the API, and then hit the part the guide does not cover: an orchestrator that fans out is useless without a backlog of real work to fan out over.
SDD isn't about managing AI agents, it's about managing context
Spec-driven development reads like a methodology for controlling AI agents. It isn't. It's a methodology for managing context across stateless sessions. The spec is the persistent memory.
Specs in, deploys out, no keyboard
Two production sites, a blog, and two personal AI projects, shipped this week from a phone. The chain is voice dictation into Perplexity Computer, a spec, then Claude Code on the web. The interaction model is the story.
Follow the work
New tools and writing as they ship — pick a channel.