Caleb Keller

In the software world, we're watching AI ruin companies and build others at the same time. Terms like "Minimum Viable Product" and "First to Market" mean less when anyone can build fast and build well. So if technology keeps getting easier, what actually differentiates? Passion and effort.

AI is changing the game. It's ruining companies and building others at the same time. The technology side is only going to get easier. So what do we need to bring? Passion and effort.

Most of what I've built was just for me. Life calendar, generative art, lame data. Even if I never tried to make anything of it, still has value. Sometimes the thing you build for yourself proves to have value for others. That's what drives me to build open code.

It's okay to just build for yourself. And AI makes that easy.

There's a lot of fear sprinkled on my feed, sprinkled with a little bit of excitement Software Engineering is dead. Data Analysis is dead. I accept that my professional worlds will never be the same, but I am not afraid. I am excited. I’m most excited by the enablement AI offers: allowing those with deep passions and knowledge to build what they previously didn’t have the skills for. Are you building something you're passionate about?

@ImAVibeCoder It actually depends on the platform they've used. I also see a lot of the boxy apps with fat shadows and that mustardy color that are from a different vibe-coding program.

Every vibe coded site has the same purple gradient and rounded cards. You know what my e-commerce sites looked like? Ugly. Still did 50M€. Ship the purple gradient. Redesign when you have users.

@PastaPizzaMario @gregisenberg I think we'll see the rise of "super apps" released by the big 3. Apps that become whatever you want based on usage patterns or simple requests.

Not to be a downer but in a world where AI apps can crank out these custom apps … we would no longer need the custom apps… The 2-3 large AI companies and their models will fill the gap that all these vibe coded apps are attempting to fill. Any successful app moving forward will be downloaded out of novelty like flappy birds , used for a few weeks then forgotten.

how to make your ai app not look like it was made with ai (weavy ai, claude, figma etc) use weavy ai to generate custom assets, color systems, and logos across models like flux + ideogram, then compose it properly in figma then paste into claude code/google ai studio etc watch

This keeps happening: Building dashboards to inform user decisions-only to immediately automate the actions the user was taking once they looked at the dashboard. Is it a problem to be building things you know will likely be replaced very soon? Often times it's a necessary first step. I'd rather be automating after building the dashboard than not having the dashboard get any use at all.

I built a small piece of software that I am actively trying to eliminate the need for. If I genuinely moved my workflow into Discord/Slack, I wouldn't ever be cd'ing into a directory. Have you ever noticed you were writing code you planned to obsolete very soon?

Data practitioners: We can't let the other half of IT move faster. It's a race, I'm competitive. We have to keep up or even pull ahead. My challenge to you: demo a vibe-coded solution. Tomorrow.

Can AI finally kill the dashboard? The next wave of BI won't look like dashboards. The next wave will be driven by words not by SQL. It will be fast and agile. It will be agentic: where a question is - researched - cataloged - answered - interrogated all without a human. And it will still all end up in Excel.

@valsopi I’m all here for personal AI, but probably not implemented exactly like that.

@calebmkeller 😅😅 It’s nothing on here. You should see twitter screenshots on insta and fb. And then the comments. looool.

The amount of the absolute bullshit on here is making me snarky and I don’t like it one bit.

@valsopi So you’re not convinced that they’re actually plotting our demise, creating new religions, and talking about us in private forums. 😅

@calebmkeller I don’t mind OpenClaw (the thing) but the sheer amount of people playing the algo and talking nonsense around it, ruins it all.

@finereli Have you tried Devin from Cognition? It has file browser, terminal, and web browser so you can see what the agent is doing in each.

It's very silly but one of the reasons I prefer Cursor to Claude Code and friends is that I can't orient myself without seeing the list of files on the left and the current file in the middle. That's especially true for the plans that Cursor generates as I don't look at code that much anymore.

It's time to go rogue. It's time to abandon the "approved" BI tools and build out your own personalized agentic approach. Put Claude Code and Codex to work. Your productivity will improve. Your end users will be pleased. Try it this week.

@finereli Is this countered at all by more people who are building?

I noticed that the rate at which I generate ideas seems to be increasing faster than the rate at which ideas can be implemented even with all the AI breakthroughs we're seeing. I think there might be infinite demand for things to be built that is growing much faster than whatever we can build with whatever models we have.

PowerBI can't do this. I can spin up a dashboard, complete with data pipeline, SQL queries, and advanced filtering with only a sentence. I can add data quality controls with another, and alerts and direct webhook calls with a third. Can PowerBI do that? That quickly? No, it cannot.

Data analysts, wake up. Traditional business intelligence (BI) tools are dying. The faster you abandon them and look to agentic tools the better. If the data world stays stuck on PowerBI and Tableau: they're certain to fall behind! Don't be one of them.

I'm noticing a disturbing trend: the ones talking about "agentic" anything are the software engineers. The lack of data folk talking AI on my feed has me worried. Is data becoming the "slow" department? Data people, where are you?