Ilya Tkachov
@ilyatkcv
Building @wispbitai - code review that fixes tribal knowledge
Here is @wispbitai enforcing codebase standards in action. I took an issue from the @posthog repo and fixed it with @cursor_ai. Cursor and other IDEs don't always follow rules, so wispbit reviewed the code and caught an issue with inline imports + duplicated code. It fed the…
code review is the capacity startups need to build in the ai era code generation is cheap. you must get good at comprehensive code review or you’ll be shipping buggy, slow software with hidden security nightmares
Here's wispbit enforcing that migration files be written a certain way Classic example of tribal knowledge that can be automated to reduce repetition in code reviews

We take care of this bottle neck but in code reviews. You can prevent a lot of problems before something makes it to QA by having extremely high codebase quality standards. A more uniform codebase = a quicker codebase to work in
QA is about to become a huge bottleneck in software development. AI lets us churn out HUGE amounts of code extremely fast, but you still need to make sure it works. AI can help with testing too, but if you _really_ want to be sure it works, you need the reassurance of human eyes…
The current state of code review tools out there don’t fix what engineers struggle with today
This has already hit our team Even with automated PR review tools, we struggle to code-review, QA and accept code fast enough We are discussing bringing back full-time paid QA to help accelerate this - something I haven't seen in over a decade
Code review is the new bottleneck and wispbit solves it by fixing tribal knowledge
Hope everyone likes doing code reviews and QA as that’s where we are headed
you can do all of this with wispbit but in just one check
how many checks is too many
ran the wispbit code reviewer using claude code with the wrong params and claude code fixed it

Copilot is deprecating custom rules within the Github UI and moving them to the codebase. This is a huge mistake. Rules should be stored on the cloud.
Just add one more rule to CLAUDE.md bro just one more rule it’s fine I swear it works bro
"How Many Instructions Can LLMs Follow at Once?" In this paper they found that leading LLMs can satisfy only about 68% of 500 concurrent instructions, showing a bias toward earlier instructions.
being generous with your ideas and secrets has always benefitted me it’s one of the reasons why i don’t believe in stealth startups you’ll be surprised how much people give back when you open up
Founders are either really good at fundraising or really good at product, but rarely both
🚀 Why We're Passionate About Rewarding Canada's Risk-Takers: Our Latest Build Canada Memo As someone who's spent years in the trenches of venture capital at Ripple Ventures, I've seen firsthand how outdated policies are stifling Canadian innovation. It's heartbreaking to watch…
Canada's outdated capital gains policies are driving entrepreneurs and investors away. We need competitive tax reform to keep talent and investment here, building the businesses of tomorrow. Today, Canada has two policies to encourage builders to start businesses the Lifetime…
self-investments always pay off, just not always how you expect
i see wispbit as the final unlock for background agents. there are teams out there that have Devin ship more than 50% of their PRs- imagine when we get to 70, 80, 90%. the final boss, the “chef’s kiss” before deployment- that’s what we’re solving at wispbit!
We’re the feedback layer for your background agents. When you run coding agents, someone should be there to review the code. wispbit does that.