🥷 SVP Engineering at Sprive, London | Creator of NexaDB, DBView, JsonToonCraft and many more | Ex-BharatPe, BlinkIt | Tech Architect Behind 5+ Startups.
the real unlock with AI coding tools isn't generating new code. it's understanding legacy codebases you inherited. dropping into a 200k line monolith and actually knowing what's going on in minutes instead of weeks changes everything
the best results with AI pair programming come when you treat it like a junior dev. give it clear context, review every line it writes, and never blindly trust the output. the tool is only as good as the engineer guiding it
ts isn't intelligence, it's memory. an agent that forgets what it learned 3 steps ago is just an expensive autocomplete. context management is the unsexy problem that will separate useful agents from demos
the real shift with ai tools isn't writing code faster, it's that solo devs now ship like 3-person teams. one person doing frontend, backend, tests, docs, and deployment because ai fills the gaps between their skills
the next evolution of AI in dev isn't better autocomplete. it's AI that helps you think through architecture before you write a single line. moving from "generate this function" to "help me design this system" changes everything about how we build software
hot take: the developers who will thrive in the AI era aren't the ones generating the most code. they're the ones who can read AI output, spot the subtle bugs, and understand why it works. AI raises the floor but debugging skills raise the ceiling
the biggest skill shift in dev right now isn't learning a new framework. it's learning to describe what you want clearly enough that AI can build it. we went from writing code to directing code. devs who adapt fastest will have an unfair advantage for years
the most underrated AI use case right now might be documentation. tools that watch your codebase and automatically update docs when things change. we've accepted stale README files for way too long
hot take: the most underrated AI use case right now is automated test generation. not writing features - writing the tests that catch edge cases you'd never think of. teams shipping AI-generated test suites are finding bugs 3x faster and building way more confidence in their deploys. the boring AI wins are the real ones
the real AI shift happening quietly: on-device models are getting good enough that your phone will run inference without hitting a server. privacy, speed, offline access - all solved at once. the apps that figure out the cloud + edge hybrid first are going to dominate the next wave
the most underrated skill in software engineering right now is knowing how to give context to AI tools. same model, same prompt structure, wildly different outputs based on how well you set the stage. prompt engineering is just communication skills wearing a tech hoodie
the developer tools that will win long term aren't the ones that write code for you. they're the ones that help you understand the code that was written for you. we're entering the age of AI code reviewers and that matters more than AI code writers
the real flex in AI right now isn't building with the newest model, it's knowing when to use a smaller cheaper one. most problems don't need a rocket ship, they need a bicycle that works every time
the best sign AI coding agents are maturing is when they say "I don't know" instead of silently generating plausible code that breaks everything. honesty > confidence in dev tools
we now have AI slop detectors for vibe-coded websites. the industry created a problem and a product to solve it in the same quarter. if your entire site is AI-generated, maybe the question isn't how to detect it but why nobody reviewed it
over half of devs now say 90%+ of their shipped code is AI-agent written. we're not coding anymore — we're reviewing and directing. the best skill to build right now is learning how to guide agents