Abhishek Bansal

Co-Founder, ProdE AI.

Plan - Spec - Ship. From PMs to Devs to QAs.

[Abhishek Bansal @bansalab93]

Claude Opus is getting more & more human!!

Giving superficial answers!
Asking users to confirm in code themselves!

And now, playing defensive and shifting the blame!

[Nilesh Bansal @prode_ai]

LLMs come and go, but your context is forever.

[Abhishek Bansal @bansalab93]

Hot take: most AI coding tools fail at planning - too quick, too shallow, never asking how deep to go. @garrytan's GStack scope calibration is the first time I've seen an AI ask 'how ambitious should we be?' before writing a single line. Every tool should do this.

[Abhishek Bansal @bansalab93]

Does it help to curse the LLMs? In my experience
- Either they are thick skinned, apologize and continue to make mistakes
- Become a yes man, and switch off their brain

[Abhishek Bansal @bansalab93]

Because of Cursor's Composer 2 release, I can rest easy that Anthropic is not going to increase its subscription prices in the near future.

[Abhishek Bansal @bansalab93]

@JsxsProductions Them capturing other shortcuts like cmd+K, cmd + shift + L is what made me switch to Vscode.

[Abhishek Bansal @bansalab93]

@rafaelobitten @OpenAIDevs @claudeai Wondering the same at an IDE level. With Cursor's composer 2 releasing, should switch back to cursor from Claude Code. But then, Anthropic will come up with a better model as soon as I get comfortable with Cursor.

[Abhishek Bansal @bansalab93]

@joni_vrbt Signal.

[Abhishek Bansal @bansalab93]

When have you been most frustrated when working with other dev teams? For me, it was when I had to consume responses from another team's APIs and the response schema kept changing with no documentation on when or why the schemas changed. #SoftwareEngineering

[Abhishek Bansal @bansalab93]

@CodeWizard For enterprise, leverage your existing network!

[Abhishek Bansal @bansalab93]

Interested in hearing how engineering leaders are approaching coding agent adoption? Are you enforcing a single agent in your org, or does each dev use their own preferred one? #engineering #cto

[Abhishek Bansal @bansalab93]

@mchulet https://t.co/l5f67dLSdE spec driven development for enterprises.

[Abhishek Bansal @bansalab93]

@Priya_Upadhyay_ This! I've had multiple instances where I had to lay out the logic for agents, otherwise they complicate things unnecessarily. Even Opus 4.6 is guilty of this, even with simple if else conditions sometimes.

[Abhishek Bansal @bansalab93]

@DnuLkjkjh Same. Especially with non-deterministic stuff like agents, you don't know what will work.

[Abhishek Bansal @bansalab93]

What workflows work best when you're just experimenting, working on a PoC, or the solution isn't clear yet? Spec driven development is great, but for experiments I have to rely on plain old iterations with Claude Code. #specdrivendevelopment #sdd

[Abhishek Bansal @bansalab93]

@AntonEvgrafov Agreed, but exhaustive planning initially can save a lot of mid project meetings + updates.

[Abhishek Bansal @bansalab93]

@AlexEngineerAI We dogfood ProdE heavily. Product, UI/UX, and engineering specs are generated in ProdE; execution happens in Claude Code.

[Abhishek Bansal @bansalab93]

@kritikakodes LLM outage!

[Abhishek Bansal @bansalab93]

Coding agents don't fix dysfunctional teams. They expose them faster.

No clear requirements.
No edge cases discussed.
"Just build this, we'll figure it out."

Now your dev finishes in 10 min and waits 3 days for someone to answer "is this in scope?"

The bottleneck was never the code.