This triggered a obra/superpowers brainstorm which asked me:
Q: Which implementation language?
A: You decide the language (it picked Python)
Q: How should we verify it actually works before declaring done?
A: Do whatever you think is best. (it picked TDD)
Q: Do you have Slack credentials ready to test against?
A: We’re going to rebuild the existing thing using a WebHook. Have an .env file and a docker compose.
Q: Webhooks can’t upload local files. How do you want to handle the image?
A: The selected file is moved to meme/already-sent. Part of the docker-compose is a service that services these files. The url is also in the .env. (it picked nginx)
It created the spec, I skimmed it, fixed one wrong assumption about the folder structure and it built a working version after. Then a few follow up prompts so that I could deploy it on my existing infra.
So What Changed?
- Agentic: No copy pasting, Claude creates all the files, wires them up and tells me how to configure and run it
- Workflow: Skill frameworks guide Claude so that it doesn’t just start building something
- TDD: Probably a bit overkill for something this simple but everything that could be tested, was tested
- Tooling: Can actually run bash, docker, python commands
- Human steers: Claude can’t know how my ISP router forwarding is set up but it wired everything correctly following my instructions perfectly
Conclusion
So the answer to “Can AI build the Slack Meme Bot in 2026?” is just: yes. A brainstorm, a quick spec, a few infrastructure-matching prompts, and an hour later the first meme was posted on Slack.
The version I built manually in 2023 took me longer with a lot more swearing.
In 2026 I’m older still, but somehow less bitter, and quite a bit less out of touch 😉