Decision: stop assigning work to Quan; escalate to Han
Current truth
quan (contract dev, Vietnam) is too slow and needs heavy coaching; made an undisclosed migration that broke the tracker. Not worth assigning more work at current pace.
Status as of 2026-06-25: both founders leaning toward ending the contract (“failed experiment”) — the second chance given after han vouched for him did not change the pattern (shipped the Codex registry-tracker update broken, apparently untested; responded to a simple newsletter-logo bug fix with a process/architecture debate instead of the fix). Cost framing: $3,500/mo ≈ 10% of monthly cash. Final call still pending.
Cultural read from the same call [idea]: the gap is startup pragmatism — knowing when to ship fast vs perfect (“the newsletter needs to go out tomorrow with logos”) — which both consider hard to hire for and counter to most org habits; remote Vietnam setup compounds it (no sit-next-to-me alignment loop).
Replacement track (2026-06-25): near-term = a capable junior/intern-level dev who reliably executes tasks (3-month retention already a win; benchmark ~€5k/mo minimum for someone experienced delivering competent work); co-location strongly preferred (Amsterdam next to Vincent, or London). Vincent to tap his Amsterdam dev network. Longer-term ambition = a proper founding engineer once website + new domain are live — profile still debated (data-science specialist vs “cracked” generalist; lean generalist: “cracked engineers do unbelievable speed and see things we don’t from a product perspective”); vision = move as a team of three, hubs in Amsterdam/London (US later), company stays US-incorporated.
First candidate in play (2026-06-28): the Amsterdam Spibel founder via Dennis — see spibel-dev-candidate; proposed next step is a paid half-day trial.
Urgency read (2026-07-03) [idea]: a short-term support dev is not a critical blocker right now, but would unlock second-tier work (ChatGPT focus, scaling experiments) that keeps getting pre-empted.
His in-flight Codex-plugins tracking work did land: demoed working dev version to Vincent on 2026-06-09 (one day before the stop-assigning decision).
Decided (2026-06-10): don’t simply end the contract — Vincent escalates to Han (Quan’s manager, replies in ~5 min): “this is taking too long, I’m not giving him more work, what’s the deal?”
Longer-term: the offshore-dev model stays IF we find a faster dev — speed is the hiring criterion, because Vincent’s capacity will inevitably disappear to projects.
Open questions
Han’s actual relationship to the agency/company — clarify.
Replacement dev sourcing — through Han or elsewhere? (Amsterdam network now the active channel.)
Final Quan call: end now, or one more structured chance? (Open as of 2026-06-25.)
Founding-engineer profile: data-science specialist vs exceptional generalist?
Timeline
[2026-07-03] (internal discussion — Vincent + Elliot) Vincent had “speak to more dev candidates for short-term support” on his sprint; downgraded to non-blocker after the prior day’s discussion — framed as the person who works second-tier priorities (ChatGPT, scaling experiments) in parallel.
[2026-06-28] (call) Vincent interviewed the first Amsterdam candidate (spibel-dev-candidate, via Dennis). Next step: discuss with Elliot, then a paid half-day working trial — Vincent’s stated hiring lesson: interview-klik has misled him ~6 times before, so trial on real work first, short contract after, longer only on a real match.
[2026-06-25] (internal discussion — Vincent + Elliot; both founders recorded the same ad-hoc call, ingested as one) Quan verdict hardened: second chance (post-Han-vouch) not working — shipped the Codex registry-tracker update broken (apparently unchecked), answered the newsletter-logo bug request with a process debate; suspicion he’s working ~1h/day. $3,500/mo ≈ 10% of cash judged hard to justify; leaning toward replacement, final call pending. Vincent’s mitigating note: the test-scenario work he assigned is genuinely hard to keep up with (Vincent changes everything every few hours) and he hasn’t reviewed it yet. Replacement strategy set (see Current truth): junior/intern executor near-term (~€5k/mo benchmark, co-located Amsterdam/London), founding engineer once website+domain live (generalist-leaning), team-of-three vision, hubs Amsterdam/London/US-later, stay US-incorporated. Vincent to tap Amsterdam dev network.
[2026-06-10] (internal discussion) Decision made at stand-up. Owner: Vincent.
[2026-06-08] (internal discussion) Concrete numbers: Vincent gives quan ~5 tasks, ~1 gets done per day; “no way with AI you can be this slow”. Vincent assigned him a scrape job for Codex plugins. Agreed tactic: track hours billed vs output (“8 hours and only this got done?”), micromanage for a few days to see if he wakes up, then raise with Han (“is this the expectation, or can you find us someone better?”) to isolate whether it’s Quan or the agency.
[2026-06-05] (internal discussion) First escalation of the concern: Vincent inclined NOT to renew Quan after this month — “way too slow”, psychologically jarring next to agent-speed work; will talk to Han. Quan says he uses Warp (a Cursor-like agentic tool). Vincent has told him the pace is unacceptable; Quan apologizes but doesn’t improve. Counter-idea raised: if he’s capable, pay for his AI tool credits/subscription (maybe he’s on a low tier).