David Kandler

Something I’ve heard consistently from 60+ eng leads at 40–100-person companies: even with mature observability stacks, the volume of actionable alerts hasn’t dropped. On-call still breaks at ~50 engineers not because of raw count, but because nothing filters what actually matters.

@pcshipp Reddit is full of experts. On everything. Employed in nothing.

X is full of builders LinkedIn is full of employees Reddit is full of...?

5/ Fine trade for internal tools. Not fine for production. The line between "vibe-acceptable" and "production-grade" is the most important call engineering teams haven't named yet. Half of them are pretending the line doesn't exist.

1/ I work at a company that catches bugs in production code. I also vibe-code half my internal tools by typing "make it work" into Claude. A confession thread 👇

4/ The catch isn't that the code is bad. It's that the code is opaque to me. I can ship it. I can't debug it. When it breaks at 11pm, I'm not fixing it. I'm prompting it.

3/ Here's what nobody on the business side admits: it's faster. Sales scripts, scraping experiments, dashboards. Vibe coding does in 20 min what careful engineering does in 3 hours. And the engineer has better things to do.

2/ Variable names: "thing", "thing2", "doStuff". Comments: "// not sure why this works but it does." If our engineering team ever sees my git history, I'm getting escorted out.

@benedictk__ would like to join as well

Germany / Switzerland - who wants to be on this?

1/8 Our paper "Evaluating AGENTS.md" just won Runner-Up Best Paper at the ICLR MemAgents workshop 🏅 We asked a simple question: do context files actually help coding agents? Two findings. Both unflattering. 🧵

Everyone's measuring AI coding ROI wrong. The pitch: AI assistants make engineers 10x faster. The reality at every Series A/B team I've talked to: PR volume up 3-5x Review quality down Bug backlog quietly compounding On-call rotations getting heavier, not lighter Senior engineers spending 30%+ of their week on triage and root-cause work that used to take 5% The bottleneck moved. Writing code stopped being the hard part. Owning code after it ships became the hard part. The "second half" of software engineering (reproduce, root-cause, fix, verify) is still 100% human. That's the gap. It's also what we're building Logicstar for: an autonomous agent that takes a bug report and ships a verified PR. No human in the loop until review. self serve is open now.

@itsaaroshi not using @logic_star_ai

I'm a vibe coder, scare me with one word.

@Samaytwt Producing code was never the hard part. Knowing which code to write, which bugs actually matter, what not to ship. AI compresses the production, not the judgment.

Unpopular opinion: "AI makes everyone a developer" is true the same way "cameras makes everyone a photographer"

The most valuable AI skill in 2026 isn't prompting. It's knowing which problems are actually worth solving. The bottleneck moved upstream. Most people haven't noticed.

@zeeg honest take: greenfield code gen is a red ocean, maintenance is a blue one. nobody wants to build the agent that owns your pager at 3am, which is exactly why that's where the real upside sits.

ngl if i ran a product that was a coding agent rn theres so many good opporunities that no one has been aggressive enough about

@Im_IrushiK Bug triage

I'm a vibe coder, scare me with one word.

@rauchg Petabyte-scale log analysis to trace one threat actor is a hell of a detection investment. The real signal here: attacker prioritized env var enumeration over direct exfil. Suggests credential chaining is the actual goal, not the initial breach.

I want to keep everyone updated on the details of the security investigation. The team performed an in-depth analysis to search for root causes and to better understand the behavior of the threat actor. We cast a very wide net, pulling and processing nearly a petabyte of logs of the entire Vercel Network and API, extending well beyond the initial Context[.]ai compromise. We now understand that the threat actor has been active beyond that startup's compromise. Threat intel points to the distribution of malware to computers in search of valuable tokens like keys to Vercel accounts and other providers. Once the attacker gets ahold of those keys, our logs show a repeated pattern: rapid and comprehensive API usage, with a focus on enumeration of non-sensitive environment variables. As a result: ◾We've deepened and widened our collaboration with partners across the industry, like Microsoft, AWS and Wiz, to further protect the broader internet. ◾ We've notified other suspected victims of this threat actor, independent of this event, encouraging them to rotate credentials and adopt best practices. We've also shipped a bunch more product enhancements. I'm extremely thankful to our team and industry partners for working around the clock. For more details on the ongoing investigation, refer to our security bulletin: vercel.com/kb/bulletin/ve…

@mitsuhiko The pricing opacity actually has a second-order effect: it makes budget conversations with eng buyers harder, they can't anchor expectations. You lose the deal before the demo.

I really want to understand what Anthropic is thinking. They are now “secretly” AB testing entry level pricing? Sorry folks, I love ya, but that is just weird.

The diff-to-coding-agent handoff is the real unlock, except "communicating nuances" is where it collapses. Coding agents normally don't fail on layout, they fail on implicit system rules no diff captures. The design language file has to be machine-queryable, not just a reference doc.

My ideal AI design tool probably something like: A canvas tool, where you can get any view of your app rendered to edit or use as the starting point for a new view. You can freely explore, duplicate, and make changes visually. You could start these renders from other tools like @linear. User feedback -> render the screen to be edited. It would have design language, system and product guidance files that help guide the overall design based on your product. Each artboard carries metadata, like the origin of the view, who created it, what changes was made when, so you could query things across your whole team. You could create areas that you want AI to fill or complete. Fill this list, complete the columns with this data or using this screenshot or something. Edits in the artboard are tracked as a diff. You export those diffs as a plan for a coding agent to build against your actual codebase. The design tool agents keep check-ins with the coding agent and try to communicate the nuances of the design so it gets built as a prototype.

Strongly agree. Building is cheap now, carrying is not.The danger is “good enough” features that linger, adding complexity, support load, and debt.Great teams ship with a sunset plan: metrics and review date from day one. If it doesn’t earn its keep, it goes.Every feature should pay rent.

@PawelHuryn This is the exact gap I’m seeing in outbound to CTOs and EMs right now. Every team pilots AI agents for code. Almost none have closed the loop on Sentry/Datadog triage yet. The on-call page still lands as raw noise for most 40–100 person teams.

The example to study is alert triage. Sentry POSTs the alert. Claude pulls the stack trace, correlates with recent commits, opens a draft PR with a proposed fix. On-call wakes up to a PR, not a blank terminal.