Engineering brief
The 12 months that rewrote the rules of software engineering
At a glance
The brief is enough
- Relevance
- Practical value
- Radar
- Time horizon
- Now
- Warnings
- None
In the last year, AI-generated code became the default, migrations shrank from years to weeks, and code reviews became theater. Engineering leaders are quitting.
Software engineering workflows are being reshaped faster than most organizations can adapt, creating risk and opportunity.
Watch if you are deciding how to restructure engineering teams for AI-assisted workflows.
Summary
The new default. Nobody writes code by hand anymore. AI generates the majority of code, and engineers routinely run five to ten agent sessions in parallel. IDEs are being replaced by agentic harnesses; Cursor’s IDE is already treated as a legacy product. The last 12 months have fundamentally rewritten how software is built.
The migration acceleration. Large-scale rewrites that once took years now finish in weeks or months. Airbnb migrated its test suite in six weeks, Uber moved from JUnit 4 to 5 in four months, and OpenAI is 90% done moving its API from Python to Rust in under six months. AI writes the bulk of the code and tests, making these migrations viable.
The broken review process. Agent-generated PRs have exploded – GitHub reports a 10x increase this year. Human code review has become theater, with developers rubber-stamping changes. Teams like Ramp are phasing out reviews for non-critical code. The tradeoff: faster shipping but more paper cuts and reliability issues across production.
The leadership exodus. Engineering leaders are quitting in droves. Causes include founder-mode pressure, demands to double output with AI, equity losses, and slow-moving organizations that refuse to adopt new workflows. Many are taking career breaks or turning to fractional CTO work, leaving a gap in experienced management.
The CPU crunch. Beyond GPU shortages, securing CPU capacity for normal workloads has become a six-month ordeal. Cloud providers in some regions cannot reserve CPUs for midsize customers. This limits cloud-agent strategies and adds a hard operational constraint.
What still holds. Teams remain the unit of work. Planning and testing are as essential as ever – AI writes most code but the ratio of test writing to code writing hasn’t changed. Non-engineers still do not ship production code. The advice: stay hands-on, build agent infrastructure, and accept that people management expectations are shifting.
My take
The acceleration is real, but the celebratory tone skips the cost. Broken code reviews and quality degradation will compound. Teams that invest in eval and validation pipelines now will have a durable advantage. The CPU shortage is underappreciated – cloud agent strategies will hit a wall without capacity. The leadership exodus is a canary; if experienced managers leave, orgs will struggle to navigate this transition. Staying hands-on is sound, but ignoring people management risks a generation of IC-first managers who can't scale.
What to do with this
- Audit your team's code review process and consider trust-based reviews for low-risk changes.
- Start a small eval pipeline in CI to catch regressions from AI-generated code.
- Discuss CPU capacity planning with cloud providers if you are exploring cloud agent harnesses.
Watch the video
This video is blocked due to your privacy settings. To watch this video, please accept YouTube marketing cookies.
Related breakdowns
Why agentic design still needs the human sketch
Design engineer Maggie Appleton on why sketching remains essential even as agents accelerate UI prototyping, and why multiplayer alignment is the next big…
AI code generation is cheap. Validation is the new bottleneck.
The real AI bottleneck isn't generating code but validating it when no human reads it. Teams must build trust through testing and observability.
Dark AI Factories Break: Context Engineering and Slow Loops Work
Dex Horthy explains why his lights-off coding factory failed and how context engineering plus incremental loops are the pragmatic path to reliable AI code.
Get TL;DW
Too Long; Didn't Watch.
A concise breakdowns of the AI and devtools videos that actually matter for engineering leaders.
Free. Weekly. No hype.
Video and thumbnails remain the property of their respective creators. tldw.news provides editorial analysis, commentary, and discovery links to original content.