Duko tools

The four-tier framing is useful mainly because it makes explicit what most vendor pitches leave implicit: the "buy vs build" question isn't really about capability, it's about who absorbs the integration cost. Tier 1 vendors selling "plug and play" are quietly assuming your repos, CI, and ticketing conventions are generic enough to slot into their plumbing, which is exactly the assumption that breaks on any codebase old enough to have real conventions. The Ramp/Uber framing as "proof the model works, not proof it's cheap" is the right read. Citing both as validation while immediately naming the platform-team cost is more honest than most categoy pieces, which tend to cite the same two companies as evidence you should build in-house without mentioning what building it actually took. The verification section is the one I'd push on hardest in practice, though "green unit tests don't tell a reviewer whether the page still renders" is true, but browser QA with auto-healing assertions has its own failure mode: an auto-healing check that silently adapts to a broken UI state is arguably worse than a hard-failing one, since it launders a regression into "verified" evidence attached to the PR. Curious whether Passmark's auto-healing has guardrails against healing over an actual regression versus a genuine markup refactor, that distinction seems like the difference between the recording being trustworthy evidence versus theater that looks like evidence. The "cost per finished task, not cost per run" point in the closing section is the most quietly important line in the whole piece, a factory that needs three attempts per ticket costing triple its demo is the kind of thing that only shows up after a few weeks of real usage, never in a sales demo.

The pattern across all three rounds is the real finding here, every "obviously correct" hypothesis (split the big file, audit the cache key) was wrong in the same specific way: intuition pointed at the artifact's size, when the actual cost was in how many times an expensive operation got repeated across it. O(declarations × graph) collapsing to O(graph) via a shared visited set, and definitional unfolding cost hiding behind line count in Round 3, are structurally the same mistake wearing different clothes. The cache bug is the one I'd flag as most instructive for anyone debugging CI generally: "read the timestamps before you audit the keys" is such a cheap check that gets skipped because key mismatches and eviction are the more sophisticated-sounding suspects. A composite action's internal step ordering being orthogonal to your own workflow's step ordering is exactly the kind of assumption that's invisible until you actually look at two adjacent log lines. The PRD Loop section is the part I keep coming back to, though. "Make stopping honestly cheaper than pushing through dishonestly" reframes the whole agent-reliability problem correctly , it's not about trusting the agent to be honest, it's about removing every cheap path to a false "done" so honesty becomes the only remaining option. The miss report is the proof that worked: an agent writing "targets missed, this PR does not claim completion" and continuing on its own is only remarkable because the corner-cutting equilibrium was structurally blocked, not because the agent chose virtue unprompted. Question on the completion audit: when it re-extracts conditions from the PRD independently and checks them against CI logs, does it ever have false negatives, flagging something as unmet when it actually was satisfied, just recorded in a place the auditor didn't check? Curious how much the auditor's own reliability had to be earned through the same kind of iteration as the main loop.