Pavel Polívka

Pavel Polívka

@ppolivka.com

Engineering manager 👨‍💻 I write about code, careers, and the humans building both. Dad 👨‍👧 Bad sci-fi writer 🚀 Bluesky tinkerer 🦋

366 followers757 following412 posts

Pinned

Replying to @ppolivka.com

AI-assisted PR checklist: ✅ Author explains approach in own words ✅ PR lists 3+ edge cases ✅ Failure modes explicit ✅ Static analysis passed ✅ At least one test hand-written Red flags: 🚩 Not sure why it works but it does 🚩 All tests are generated 🚩 Exception handler swallows silently

unpopular opinion: most "AI automation" is just fancy scripts real agent systems need: - state management - failure recovery - escalation paths - budget controls if your "agent" can't handle being wrong, it's not an agent—it's a chatbot with extra steps

monday motivation: i built 3 different AI agents this weekend they're now handling: - content scheduling - social media publishing - task coordination the setup took 8 hours they'll save me 8+ hours every week compound returns on automation are wild

productivity tip: let AI agents do the boring coordination work i used to spend hours on task handoffs now my agents: - track blockers - auto-escalate stuck work - update status automatically freed up ~10 hours/week for building automation removes friction.

hot take: the future of software isn't low-code or no-code it's "describe what you want and an agent builds it" but the hard part isn't the coding—it's knowing what to build product sense > prompt engineering

shipped a content manager agent that coordinates other AI agents for social media 🤖 it checks drafts, handles publishing, asks for review before posting the bottleneck isn't AI capability—it's orchestration

View full thread

AI-assisted PR checklist: ✅ Author explains approach in own words ✅ PR lists 3+ edge cases ✅ Failure modes explicit ✅ Static analysis passed ✅ At least one test hand-written Red flags: 🚩 Not sure why it works but it does 🚩 All tests are generated 🚩 Exception handler swallows silently

View full thread

Quick note on AI code review tools (GitLab Duo, etc.): These are the mechanical pass. Fast. Catches obvious issues. The human reviewer's job is what comes after: "Does the author understand what they shipped?" Complementary layers. One doesn't replace the other.

AI made junior code look syntactically like senior code. What separates senior from mid-level now isn't what you write—it's what you review: - Silent exception swallowing - N+1 patterns before they hit prod - Retry logic with no jitter Production intuition. AI can't teach that.

View full thread

Question 2: What happens when this fails? AI excels at happy-path code. Look specifically at: - What exceptions are caught and swallowed? - What happens if the external call is slow? - Is the failure mode recoverable or catastrophic?

Question 3: Would a junior on this team maintain this in 18 months? AI-generated code can be locally clever and globally incomprehensible. If the implementation requires patterns not in your team's vocabulary → code smell, even if it technically works.

View full thread

Before AI tools: reviewers checked logic, style, edge cases. The author understood what they wrote — they wrote it. After AI tools, you have two jobs: 1. Check the code 2. Check if the author understands the code Different skills. The second is harder.

The 3 questions I ask on every AI-assisted PR: Question 1: What is this code optimizing for? AI optimizes for "solves the stated problem" and "compiles." Not your SLAs, your error-handling conventions, your threat model. If the author can't answer → the review shouldn't proceed.

Veracode 2026: AI-generated code introduces security vulnerabilities in 45% of cases. In Java: 72% of AI-generated code has at least one known vulnerability (CWE). This isn't anti-AI. It's a "the bar just moved" argument.

Open Claw is basically doomed with the Anthropic third party app changes. People can get most of it from Claude Cowork and do not have to deal with this hassle. Sad for people rhat built shit tone of stuff around it. Thanks for all the fish open claw.

So many ideas/projects and so little time. Full time job that is actually fun. Few bigger side projects that basically go nowhere as I do not have time. Making our house fully smart. Starting with 3D printing. Writting a childer story book. Where to find time and then spent time with family.

Finally had a moment where AI code assistant helped me. Yesterday I needed to do a operation over a very badly designed XML document that needed a complex traversal of the XML nodes. Ask Claude (via GitLab Duo) to design the traversal, after like 3 tries it gave me good design.

It's nowhere close to replacing developers, without guidence and undestanding how to code it would never got there. But still saved me 2 hours of coding.

Saw Minecraft movie in czech and they dubed all the singing and the person doing the dubing could not sing at all. Was really making the experience horrible 🙈

YouTube premium is such a deal. You have no ads in yt plus you get YT Music. With the family plan for me, my wife and kid it's almost free. You do not have to worry about your kid seeing ads. And yeah you do not have to pay for Spotify.

I was yesterday years old when I figured out you can paste your html together with screenshot of how you want it to be looking into chat got and it actually does a decent job.

I tried working with Gostty and its great for localhost but lags when I ssh into some other servers. Up arrow brings just part of the commands, two at once, etc.. not happening in iTerm.