๐จ Why Most Engineers Arenโt Ready for AI-First Engineering
Originally published on LinkedIn
๐จ ๐ช๐ต๐ ๐ ๐ผ๐๐ ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ ๐๐ฟ๐ฒ๐ปโ๐ ๐ฅ๐ฒ๐ฎ๐ฑ๐ ๐ณ๐ผ๐ฟ ๐๐-๐๐ถ๐ฟ๐๐ ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด A former colleague asked me to speak to engineering managers about โAI-first engineeringโ so thought where to start ๐ค
The biggest blocker isn't the technology, itโs that most engineers aren't trained with the skills AI-first engineering actually demands.
Management are now being asked to retrain teams that have worked the same traditional way for years, mainly only coding for many.
AI doesnโt just make engineers faster, it changes the bottleneck from implementation to:
๐น architecture
๐น test strategy
๐น technical communication
๐น orchestration
๐น product thinking
๐น decision making
๐ง๐ต๐ฒ ๐บ๐ผ๐ฟ๐ฒ ๐ถ๐บ๐ฝ๐น๐ฒ๐บ๐ฒ๐ป๐๐ฎ๐๐ถ๐ผ๐ป ๐ด๐ฒ๐๐ ๐ฑ๐ฒ๐น๐ฒ๐ด๐ฎ๐๐ฒ๐ฑ ๐๐ผ ๐ฎ๐ด๐ฒ๐ป๐๐, ๐๐ต๐ฒ ๐บ๐ผ๐ฟ๐ฒ ๐๐ฒ๐ป๐ถ๐ผ๐ฟ ๐ฒ๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด ๐ฐ๐ฎ๐ฝ๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ ๐บ๐ฎ๐๐๐ฒ๐ฟ๐.
Here is a short list of the big skills any engineer needs to use AI effectively.
๐ง ๐ฆ๐๐๐๐ฒ๐บ ๐๐ฒ๐๐ถ๐ด๐ป AI agents only work well if you can clearly describe architecture, boundaries, patterns, and non-functionals for someone else to build. Learning to communicate technical ideas without coding is learnt later in an engineers career only when they start delegating coding tasks.
๐งช ๐ง๐ฒ๐๐ ๐ฆ๐๐ฟ๐ฎ๐๐ฒ๐ด๐ If you manually inspect every generated line of code, you become the bottleneck immediately. The real skill becomes designing validation and safety rails. This requires forward thinking of test approach and designing testable software.
๐๏ธ ๐๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฎ๐น ๐ฅ๐ฒ๐๐ถ๐ฒ๐ Youโre no longer reviewing syntax. Youโre reviewing decomposition, boundaries, operational correctness, and system design. Reviewing AI generated code is not about code, its more akin to an architectural review, tests ensure the code actually works, this is about reviewing the delivered output design.
๐ ๐ ๐๐น๐๐ถ๐๐ฎ๐๐ธ๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ข๐ฟ๐ฐ๐ต๐ฒ๐๐๐ฟ๐ฎ๐๐ถ๐ผ๐ป This can be hard for any engineer, its not linear work. AI-first engineering is highly parallel work. Multiple agents, designing next item, multiple PRs, continuous integration, continuous validation and UAT.
๐ ๐ง๐ฒ๐ฐ๐ต๐ป๐ถ๐ฐ๐ฎ๐น ๐๐ฟ๐ฒ๐ฎ๐ฑ๐๐ต You don't implement every technology choice yourself, but choose the wrong stack, wrong provider, or pattern its massively expensive later.
๐ ๐๐๐๐ถ๐ป๐ฒ๐๐ ๐๐ผ๐ป๐๐ฒ๐ ๐ Implementation becomes quick, deciding what give value becomes more important than building it.
This is the problem:
AI-first engineering assumes an unusually senior skill profile yet all engineers overnight were expected to use AI.
A junior engineer can generate a function with AI. That does not mean they can design, validate, orchestrate, and evolve a production-scale system.
๐๐ ๐ถ๐๐ปโ๐ ๐ฟ๐ฒ๐ฝ๐น๐ฎ๐ฐ๐ถ๐ป๐ด ๐๐ฒ๐ป๐ถ๐ผ๐ฟ ๐ฒ๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐. ๐๐โ๐ ๐ถ๐ป๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ป๐ด ๐๐ต๐ฒ ๐ฝ๐ฟ๐ฒ๐บ๐ถ๐๐บ ๐ผ๐ป ๐๐ต๐ฒ๐บ.
The question to is how do you upskill your engineers? How many in the team actually have those skills?