Andrei Kalamkarov

I lead web in higher ed and spend too much time experimenting with AI tools. Writing about software, work and what I notice. Husband, dad.

[Andrei Kalamkarov @kalamkarov]

@getpantheon Really like the focus on shared context and review workflows. How do teams maintain that context over time, especially the reasoning behind past decisions? Is that managed in P1 or brought in from existing documentation?

[Andrei Kalamkarov @kalamkarov]

@cmyharish Agreed on release controls. Pantheon's take seems to be that agents get no shortcut to prod, same review as everyone. I'd guess the harder part across 25 properties is the reviewer keeping up when an agent ships 40 changes a day.

[Andrei Kalamkarov @kalamkarov]

I’m trying hard to love Grok Bot. There’s a lot to like! But I asked whether it had given me some specific advice, and it said no, several times. I scrolled up and found the exact exchange literally from yesterday.

[Andrei Kalamkarov @kalamkarov]

@ConnorTalksAI What have you been hearing? In my experience these products fluctuate in quality quite a bit, and I suspect Grok Bot is hitting some temporary infra issues that would explain it. I'll keep using both though!

[Andrei Kalamkarov @kalamkarov]

@karpathy @omarsar0 Apparently I count as a dinosaur (GitHub Copilot beta tester, 2022). I recently prepped a talk on AI dev tools for colleagues. A few days later, when I gave it, parts of my slides were out of date. Makes me wonder how we train people who don't have time to constantly experiment.

[Andrei Kalamkarov @kalamkarov]

@Grady_Booch This resonates with what I've noticed. AI removed a lot of the grind, but I didn't expect to miss some of it. Repetitive work was sometimes a mental break. Now the days are filled with decision after decision and constant context switching. It's harder to get into a focused flow.

[Andrei Kalamkarov @kalamkarov]

Using AI means less grind, but some of that grind was a kind of mental vacation. Now my days are decision after decision, and there's no real break.

[Andrei Kalamkarov @kalamkarov]

To be specific, I've been using Fable on Claude and Astra on Codex for most of my work.

[Andrei Kalamkarov @kalamkarov]

I’ve been using Claude Max 5x for a long time and recently started using ChatGPT Pro as well. I was surprised that Pro removes the 5-hour limit, but the weekly allowance still drains pretty quickly. After two days, I’m already down to 30%. I feel Claude Max 5x goes a lot further.

[Andrei Kalamkarov @kalamkarov]

An annoying thing I do often when working with AI is take a screenshot and drag it from my desktop into the terminal.

Maybe I’m late to this, but adding the Control key to Command + Shift + 4 copies it straight to the clipboard. Then I just paste it. No dragging needed.

[Andrei Kalamkarov @kalamkarov]

Crazy how much a font can affect my attachment to a product. I love Claude Code in Warp, but lately I’ve been using Codex for a lot of tasks and something just felt off.

I changed the Codex font to more closely match my Warp setup, and suddenly I like Codex way more!

[Andrei Kalamkarov @kalamkarov]

Surprised that Claude refused to recommend a specific investment for my self-directed account, while GPT and Gemini had no problem doing so with the same context.

[Andrei Kalamkarov @kalamkarov]

Getting photos off my Pixel onto a Mac was a struggle for years. The connection didn't always work with Android File Transfer or OpenMTP, or died halfway. My go-to fix was restarting either device and hoping it would work.

Now the best way is to ask AI. Super reliable and fast.

[Andrei Kalamkarov @kalamkarov]

Wrote about knowing when to stop: https://t.co/Q9caA3qyu9

[Andrei Kalamkarov @kalamkarov]

AI removed the part where getting tired used to make me stop.

That's the annoying thing: it can make a bad habit feel productive. Now I can polish something past the point of usefulness and call it iteration.

[Andrei Kalamkarov @kalamkarov]

I love building slide decks with AI. It's so powerful.

But the speed of iteration is a real bummer. I'm refining text and layout together, the whole picture, and every change takes minutes to land. That kills the rapid fire brainstorming I actually want.

[Andrei Kalamkarov @kalamkarov]

A few days ago I posted that I apparently operate in YOLO mode personally, while keeping strict "ask" and "plan" guardrails on my AI work.

Turns out that's the whole point: AI safety needs more than rules. It needs habits and boundaries.

Wrote it up: https://t.co/7WvRqHsbMS

[Andrei Kalamkarov @kalamkarov]

Deleted something today in a rushed decision. Realized it was a mistake shortly after. Entirely my fault. No AI involved.

Funny, that's exactly why I set strict guardrails with "ask" and "plan" modes for all my AI work.

I apparently operate in YOLO mode.

[Andrei Kalamkarov @kalamkarov]

@andrewchen Less religious than emacs vs vim, I think. Long-time Cursor user, also on Claude Code and Codex, and I always drift to the terminal even when I start in Cursor. Same reason I'm on Cursor CLI now. Snappier, more focused, more autonomous. The preference is clear.