EUWardenAI SDLC · results
An evidence-based process — so here is the evidence

Results from a working process

Every number comes from our own, day-to-day delivery — the VaultPAM security product, built in the AI SDLC by a team where most roles are filled by AI agents. Sources are given next to each number; methodology at the bottom of the page.

As of 26 Aug 2026 · sources: GitHub (pull requests) and Jira (tasks) · ~12 months of delivery
01Pace

Code stopped being the bottleneck

6,162
merged pull requests since the process started
GitHub · PRs merged, full period
791
PRs merged in the last 30 days — about 26 per day on average
GitHub · PRs merged, 27 Jul–26 Aug
0.6 h
median time from PR opened to merged
GitHub · last 1,000 merged PRs (13 Jul–27 Aug)
93%
of PRs merged less than 24 hours after opening
GitHub · last 1,000 merged PRs (13 Jul–27 Aug)
1297 March 733 April 1196 May 1201 June 927 July 715* August
Merged pull requests per month (GitHub). * August — month in progress, as of the 26th; it is also the month of a deliberate rebuild of the process itself, run in the same working mode.
02Control

Pace does not cost control

A 0.6 h median from PR to merge does not mean "nobody is looking". It means machines are looking — immediately: verdicts, CI and gates run in minutes instead of waiting in a queue for someone's calendar.

4,979
tasks driven to "Done" (out of 5,732 created)
Jira · full period, excluding 524 "Won't Do"
529
tasks closed in the last 30 days
Jira · resolved, 30 days
68
stops at the QA gate — about 1.2% of all tasks (68 of 5,732) returned for rework before merge, each with its own ticket
Jira · QA-stop tickets / all tasks
50
defined agent roles — the core is the 11 roles described in the presentation; the rest are domain specializations, advisory and utility roles; a dozen or more active in commits every month
repository · role definitions

68 QA stops are not an embarrassing statistic — they are a working brake. Each one is a defect that never reached the main branch, has a named reason and a ticket, and the work went back for a fix along the same auditable path. See such a return step by step in the walkthrough demo.

03Methodology

Where these numbers come from — and what they don't say

sources
GitHub (pull requests and merge times, search API) and Jira (tasks, statuses, QA-stop tickets). Numbers collected 26–27 Aug 2026 (GitHub) and 31 Aug 2026 (Jira); the page is updated manually, not live. Merge-time metrics are computed on the last 1,000 PRs (~6 weeks), not the full period — they measure the current rhythm of the mature process; the early months of building it do not describe the present state.
scope
One product (VaultPAM — PAM for European companies, SOC 2 / ISO 27001 audit standards), one team, ~12 months. This is production delivery, not a benchmark or a showcase project.
what they don't say
PR count does not measure business value — it measures throughput and rhythm. Median merge time says nothing about task difficulty. We do not publish token costs or a comparison against a human team here, because we have no parallel control group — we run that comparison only in a customer pilot, against the customer's own baseline.
verifiability
On request we show these metrics live — no cherry-picking: full PR lists, stop tickets and task runs, on screen, from our systems.
04External context

Not just our numbers

Our metrics come from a single team — so we set them against independent industry research. Its conclusions are not flattering for AI, and in our view rightly so: they show what happens when speed has no process. In March 2026 DORA added the other half — small batches, automated testing and fast feedback loops as the counterweight to destabilization. Every number below can be checked at the source — the links sit under each card.

DORA 2025 · Google Cloud · ~5,000 respondents, September 2025

AI adoption now correlates positively with delivery throughput and still negatively with delivery stability. 90% of respondents use AI at work, over 80% report a productivity gain — and 30% have little or no trust in AI-generated code.

Why we cite it: this is a change from DORA 2024, where throughput was negative too. The gap did not close, it moved: speed went up, stability did not follow. Gates address exactly that second half.

CodeRabbit · December 2025 · 470 open-source PRs (320 AI-co-authored, 150 human)

AI-co-authored PRs: 10.8 issues per PR against 6.5 in hand-written ones — roughly 1.7×. At the 90th percentile AI PRs reach 26 issues per change, more than double the human baseline.

Why we cite it: agent code needs harder verification, not softer. Our 68 QA stops (~1.2% of tasks) are not an embarrassment — they are the mechanism that neutralizes these statistics before merge.

METR · March 2026 · 296 agent PRs reviewed by maintainers

Maintainer merge decisions ran on average 24 percentage points below the automated SWE-bench grader score. Control: known-good, human-written patches passed the grader 100% of the time and won reviewer approval 68% of the time.

Why we cite it: green tests are not a merge decision — for anyone, not even for a human. For agents the gap is ~24 points wider. That is why the gate verdict here is issued by a role other than the change's author.

See these numbers live Process presentation → Walkthrough demo → Adoption path → Process architecture → Process tools → FAQ →