Himakara Pieris

Founder, DeepModel.

[Himakara Pieris @himakara]

Software is valuable because it solves the right problem, in the right way, inside a system that can be understood, maintained, extended, and trusted.

If you loop your way into a software “product” that’s not well understood and is hard to maintain, extend, and trust, there’s a low probability that you’ll create something of value.

[Himakara Pieris @himakara]

Not a problem. Looping is a thing now ;)

[LangChain @LangChain]

A coding agent gets stuck in a retry loop overnight
By the morning, it’s made 10,000 LLM calls
You now have a four-figure invoice

Observability tells you what happened, but stopping these problems before they happen means enforcing policies at the request layer.

[Himakara Pieris @himakara]

We use loops with human orchestration :) this is what works for us to write production-level code. https://t.co/Dcav50n14C

Automatic phase progression (i.e., looping) is extremely ineffcient and tends to drift. I've used this human in the middle loops for like the last 18 months.

[Himakara Pieris @himakara]

Here’s the exact workflow I run with teams (5 disciplined phases, human gates at every step):

Specify → Plan → Build (phased) → Validate → Ship

Engineer stays in control. AI executes and stops. Full audit trail in the repo.

Full AI SDLC + principles + in-practice playbook: https://t.co/Dcav50n14C

Who’s already running something like this in production? Drop your workflow (or biggest friction) below 👇

[Himakara Pieris @himakara]

Engineering leaders: the shift is already here:

Old world: Software Engineer = ticket -> code -> PR New world: Product Engineer = problem -> spec -> AI agents -> outcome

AI doesn’t replace engineers. It finally lets them stop cranking code and start owning the product.

The bottleneck moved from typing to judgment.

[Himakara Pieris @himakara]

The real unlock: responsibility stays with the developer.

AI does not own the outcome. The developer does.

When the feature is wrong, when the system is unmaintainable, when security fails — the model isn’t accountable. The engineer is.

The developer owns the “what” and “why.” AI only owns the “how.”

[Himakara Pieris @himakara]

We need a new developer workflow.

A workflow designed for a world where code is abundant and judgment is scarce:

https://t.co/MI6jLfXt94

[Himakara Pieris @himakara]

It has never been easier to build software.

It has never been harder to build useful software.

[Himakara Pieris @himakara]

Many teams are using AI with a software development process designed for a world where code was scarce. That world no longer exists.

[Himakara Pieris @himakara]

In enterprise AI, adoption tends to follow procurement gravity rather than product quality

[Himakara Pieris @himakara]

You don’t win by preserving optionality. You win by removing it.