@carrynointerest So I think "you only want frontier" is largely true - basically unless you are either bootstrapped or at a mature stage.
Of course, there's the odd startup that is cost-disciplined from day zero, but common advice holds this irrational pre-validation.
@carrynointerest Early startups are often tolerant of poor unit economics because they don't have anything to scale, have tons of credit to burn, and growing beats optimizing.
Optimizing generally makes sense at higher scales, under tough macro conditions, or once you have something to optimize.
@alexanderrX_ SF makes it easier to top up the bank with well-paid jobs between riskier gigs. It's more rational to take a multi-year salary cut when 75th percentile salary is ~$390k vs. £140k.
Tempted bring back "releases" instead of CD:
- Merge PRs extremely fast and cheaply (<5m, <$1)
- Slow/costly/thorough release checks (<30m, <$10)
- Cut release trains hourly or ad hoc.
- For more autonomy, check harder & release less frequently
The Codex app is a feat of engineering. It's really hard to build reliable systems that manage such complex state universe and AI quality under extremely variable inputs. You would expect a lot more edge cases (and you see them in other harnesses).
Even just the slickness and capabilities of the embedded browser compared to a Chrome extension. They seem to be making bold choices and reaping the benefits of it.