Bojan Spasić

Software engineering and agile process coach | I help teams and companies to build better software and deliver it faster in cost-efficient ways

[Bojan Spasić @bojanspasic]

@RoseSilicon Well, surely not if you use them like that. Luckily, there is a lot more to putting rules or a layer of guardrails.
Turns out that all that tech excellence rigor that I’ve been teaching devs for years, really pays off with agentic AI… agents seem to be good at following those.

[Bojan Spasić @bojanspasic]

Programmers have raged against abstractions for 70+ years:
• 1950s: “Compilers can’t beat hand-optimized assembly. Real coders write assembler!”
• 1980s: “OOP?Just C plus overhead & leaky abstractions”
• 2020s: “LLMs hallucinate buggy code & make us obsolete”

Denial again?

[Bojan Spasić @bojanspasic]

Pushing for better prompts will get you only so far.
Using planning mode also.
Do cross-model verification ditto.

Only using AI agents in real technical excellence workflows might get you to really autonomous automatic programming system.

#GenAI

[Bojan Spasić @bojanspasic]

Building a personal website or a website for small business is solved...

[Bojan Spasić @bojanspasic]

So stop worrying about being replaced by AI.
Start focusing on the skills AI can’t easily replicate.
Level up your architecture, requirements, testing, and judgment.
The future belongs to those who know how to use AI effectively — not just those who let it use them. /X

[Bojan Spasić @bojanspasic]

You will not be replaced by AI (yet!) — as long as you still bring real value to the table.
Coding is (almost) solved by AI. But what else do you bring? 👇

[Bojan Spasić @bojanspasic]

AI is an incredibly powerful multiplier.
But it still needs skilled humans to:

- Define the right problems
- Design robust systems
- Validate quality
- Make smart trade-offs

The winners won’t be the fastest coders. They’ll be the best system thinkers and AI drivers.

[Bojan Spasić @bojanspasic]

As it was the case with any technological breakthrough. Democratization creates demand.

[klöss @kloss_xyz]

Karpathy just told us engineering job demand is still rising. Contrary to the narratives. That cheaper engineering creates more demand, not less.

The guy who helped co-found OpenAI.

And today, OpenAI announces they're nearly doubling their org from 4,500 to 8,000.

Coincidence?

[Bojan Spasić @bojanspasic]

We need to understand that no automation needs to be error-free. It is enough to be better than humans… and repeatable.

[Bojan Spasić @bojanspasic]

This is wow… just wow… 😮

https://t.co/STlaxBr6uK

[Bojan Spasić @bojanspasic]

Claude can fix them, but relying on it for every edit is annoying as hell.

Next: polish the console UI to make it actually pleasant to use, then do a major architecture overhaul.

Repo still updating live: https://t.co/G5307slfDW
6/6

[Bojan Spasić @bojanspasic]

Spoon-feeding Claude every single requirement is getting old fast. Prompts are ephemeral — stuff gets lost, no persistence.
So I switched to user-defined markdown tasks/stories in a specs/ folder. 1/5

[Bojan Spasić @bojanspasic]

Downside: code is becoming almost impossible to change manually.
The retroactive/fragile test suite (yep, rookie mistake we talked about) makes refactors scary — one tiny tweak breaks tons of tests that weren't even well-written to begin with.
5/6

[Bojan Spasić @bojanspasic]

Things to ponder:

How do we audit "emergent" code from agents?
When does coverage + mutation stop being enough?
What's the right human oversight dosage?

Repo: https://t.co/G5307slfDW

8/8

[Bojan Spasić @bojanspasic]

Continuing the "naive vibe" coding journey with White Castle digital (treating Claude as black-box dev buddy, no peeking at code myself).
Got all core functionalities in. But yeah… tech debt snowballed hard. Time to see how solid (or scary) this really is. 1/8

[Bojan Spasić @bojanspasic]

Reviewing the full growing codebase manually at this point? Unrealistic.
Even with 90%+ coverage and 85%+ mutation score — trust is still partial.
This experiment shows both the power and the blind spots of letting AI run wild(ish) on non-trivial projects. 7/8