At a glance
WHAT IT’S REALLY ABOUT
How AI shifts engineering bottlenecks, norms, and org design choices
- Engineering bandwidth and code generation are no longer the primary bottlenecks, shifting constraints toward verification, review, security, and cross-functional coordination.
- Claude Code reduced upfront planning and design-doc rituals in favor of just-in-time planning, prototyping, and using PRs as the primary vehicle for discussion and alignment.
- Technical debates increasingly resolve through generating multiple working implementations, making “code wins” cheaper than prolonged argument—while requiring strong culture to avoid ‘last check-in wins.’
- AI-assisted code review is heavily used for linting, style, tests, and early bug-catching, but human experts remain essential for risk-sensitive areas like legal, security boundaries, and product taste.
- The team intentionally stays flat and scrappy (including managers starting as ICs), continuously kills obsolete processes, and tracks success via onboarding speed, PR cycle time, and Claude-assisted commit rates alongside quality/reliability outcomes.
IDEAS WORTH REMEMBERING
5 ideasAssume your bottleneck has moved; redesign norms around the new constraint.
When code becomes cheap, teams generate more of it, and bottlenecks migrate to verification, review bandwidth, CI capacity, security, and cross-functional approvals. Reassess workflows that were built for a world where typing code was the scarce resource.
Replace long-range roadmaps with just-in-time planning and prototypes.
Six-month plans decay quickly in fast-moving AI product landscapes, so Claude Code biases toward shorter-horizon planning and shipping prototypes to internal users and customers for real feedback instead of extended pre-planning rituals.
Use “code wins” to settle debates—generate options, then evaluate impact.
Instead of whiteboarding endlessly, Fiona generates multiple PRs to compare approaches, including downstream effects on API callers. This shifts debate from hypothetical arguments to concrete tradeoffs and user/developer impact.
Double down on shift-left verification as throughput rises.
Higher output increases the chance of regressions and “new ways to break,” especially when non-engineers are also shipping code. Invest in automation and early checks so confidence stays high without slowing delivery.
Automate the question behind “who owns this code?” rather than clinging to old ownership models.
With most PRs being AI-assisted, “who wrote it?” is less informative; teams should identify whether they need accountability for a regression, an expert for support, or context for decisions—and build tooling/routines (e.g., automated summaries) to answer that faster.
WORDS WORTH SAVING
5 quotesWhat served you prior may not serve you any longer.
— Fiona Fung
For years, engineering bandwidth was the expensive thing.
— Fiona Fung
We tend to just layer more and more and more processes on.
— Fiona Fung
Nowadays, I can just generate all the different options we've been discussing.
— Fiona Fung
When building is cheap, arguing expensive, again, how does that shift your team norms a bit?
— Fiona Fung
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome