Why code review is the single most important category in software “I actually think code review is the single most important category today in software. By far the most important category. The rate at which we are building code is unprecedented, but it is never production-ready or production-intent. The... amount of time it takes to review, process it and make it production-intent is still an order of magnitude more than building. It is much more important than code generation.” Jason ✨👾SaaStr.Ai✨ Lemkin Love to hear your thoughts Daksh Gupta Harjot Gill Itamar Friedman Zach Lloydshow more

I would say “verification” of correctness of software as a whole is becoming the very important category. Code review is a trivial way to verify. But it’s not enough: lightweight formal verification, o11y + monitoring + alerting all in this group It was clear ~a year back and now it’s happening

code review / code validation are evals for your company. your code validation platform is where you define the ideal state and expected behavior of your product.

100%. Generation is getting cheaper by the week. But, the goal isn’t more code per se. It’s getting to an L4 software factory that creates meanignful business value. We all want to create meaningful value for customers and users much faster, while staying in control, even growing control especially considering the risk surface of software keeps growing. Those who do not invest in code review and governance are taking double risk: both being stuck with low quality unsafe software that humans will have hard to unstuck, and not being able to move in growing velocity that some other competitor will be able to acheive

I think the AGI-pilled take here is that agents will get good enough that “review” is no longer necessary as a distinct step in the SDLC at all.

i review more than i write now

most of my problems get caught earlier, when the agent tells me its plan before writing any code

Code review not hard with an eval agent and an LLM coordination tool like (or others) agent orchestration is the opp’y

Critical code review questions that must be addressed in order to truly ship high quality code lives outside the changed LOC: which services depend on this behavior? which assumptions are undocumented (agents love assumptions!)? does the code change still serve the business intent? More generated code increases the value of that context.

never production ready is the honest part

that's fair — review is where most bugs slip through

Sorry. Silly argument.. human code is 100x worse even at 100th of the volume.

More work checking it than making it. Finally, AI understands my internship.

Code review is the future! It sets a standard for what's acceptable or not, mate. LFG very hard.

review used to mean checking someone's thinking. now there's nobody to ask "why'd you do this" so you end up re-deriving the whole diff yourself

Review got heavier for me too once agents wrote most of the code. What helped: write requests by stage, not by recipient. A request that says "schema change" gets read. One that just says "for review" gets skimmed.

ours showed green for 8 days while reading nothing, which is how i ended up planting bugs to check the reviewer

Great point. I think what many people miss is the fact that even the best model in the world would generate just about 90% of your intent. Every generation needs verification and that’s a much harder problem, because completing a task vs. making that task fit into a large system is a very different game. Shoutout @QodoAI that understands this problem space very well.

Well code reviews become less relevant if you 1) get the coding in a loop anyways to have gone through automated testing 2) you have really good bug identification process in production 3) automated roll back process ; but yes code reviews now are important. Interestingly we were running analysis on our code reviews today and 87% of our pr reviews were ultimately pushed live with zero human comments over the past month

A review tool that puts the risky changes first would help more than one that leaves 40 comments. Auth, payments and data migrations deserve a different level of attention than a renamed variable.

I'd argue it isn't code review at this point, its implementation review. Did I get the result I wanted and expected, or not? What is needed to close the gaps? That is the where the value currently is.

Not just code review, but testing as well. (Verification layer) Two must have downstream task for any code. You review the code so that you can catch issues and system is manageable as you grow the system. And you need to test it to ensure that what it works as expected.

agreed, though i think review moves earlier. deciding what an agent has to prove before it starts saves more time than reading its diff after

Agree with @jasonlk. AI code review is an emerging area of agentic security. While cyber/frontier models can build code and identify vulnerabilities, every model is different. Challenge is prioritizing, validating, and remediating vulnerabilities from all models at machine speed.

Software architecture is more critical than ever. AI writes fast and the code “works”, but it won’t scale because fundamentally, it’s trying to achieve an immediate, short-term goal. It works great until it doesn’t.

More code is easy to measure. Reliable software that solves the right problem is the real outcome.

Agreed, and the cheap half of review is the one teams skip. A model reviewer catches logic. A plain script catches rules: banned patterns, formats, anything you can write as a check. My publishing gate is a short Python file, and tonight it failed a draft that a strong model had already reviewed and approved. When code volume explodes, the deterministic layer is what keeps review cost flat, so the expensive reviewer only sees what a rule can't judge.

Review is the bottleneck I hit running AI in my own business too. My bots can draft far more than I can read, so each one now checks its work against a written checklist first and I only look at what passed.