Hurair | AI for Builders

Breaking down AI tools, models & updates into what builders can actually use 🔧 Daily signal, zero hype. Software engineer based in Berlin.

[Hurair | AI for Builders @HurairHashmi]

Hello — I break down AI tools + model updates for builders.

What matters, what’s hype, and what you can actually ship this week.

If you’re building with Claude/GPT/Gemini/Kimi + agents + automation, follow along (and tell me what you’re stuck on).

[Hurair | AI for Builders @HurairHashmi]

What’s the one repetitive task you do every week that you know you could automate…
but you either don’t have the time or skills to build it yet and would pay a monthly subscription or a one‑time setup + monthly fee.

Drop your pain point below 👇

[Hurair | AI for Builders @HurairHashmi]

What are you building with AI right now that’s close to monetization—or already making money?
Curious to hear what you’re working on 👀

[Hurair | AI for Builders @HurairHashmi]

`/btw` in Claude Code = a “spec lane” while the “execution lane” keeps moving.

Use it for: backward-compat decisions, acceptance tests, rollback plans — then feed the decision back into the main task.

[Hurair | AI for Builders @HurairHashmi]

Rule: `/btw` is for questions + decisions.
Then you explicitly feed it back:

“Back to main task: do X based on the `/btw` decision.”

[Hurair | AI for Builders @HurairHashmi]

Hello — I break down AI tools + model updates for builders.

What matters, what’s hype, and what you can actually ship this week.

If you’re building with Claude/GPT/Gemini/Kimi + agents + automation, follow along (and tell me what you’re stuck on).

[Hurair | AI for Builders @HurairHashmi]

Use cases I’d actually use:
- btw: backward compatible or not?
- btw: what’s the rollback/failure mode?
- btw: here are acceptance tests—optimize for these

[Hurair | AI for Builders @HurairHashmi]

What it actually fixes: long runs die on ambiguity.

`/btw` gives you a “spec lane” while the “execution lane” keeps moving.

[Hurair | AI for Builders @HurairHashmi]

Claude Code just added `/btw`.

It lets you start a side conversation with Claude *while it’s working*—so you can ask quick questions / clarify requirements without interrupting the main run.

[Hurair | AI for Builders @HurairHashmi]

Builder takeaway: steal the pattern.

Even if you never run "zero employees", you *do* want:
- repeatable workflows
- enforced budgets
- inspectable decisions

That’s how you scale agents without babysitting.

[Hurair | AI for Builders @HurairHashmi]

Paperclip (open source) is getting attention as an "operating system" for running a company with agents.

Hot take: the org chart is the least interesting part. The real product is *control planes* for autonomy.

[Hurair | AI for Builders @HurairHashmi]

Most agent demos fail on 2 things: (a) state drifts, (b) nobody owns the spec.

A real "AI company" needs: explicit goals, measurable KPIs, and a human-in-the-loop that intervenes *before* damage.

[Hurair | AI for Builders @HurairHashmi]

@Saboo_Shubham_ This is the right direction — agents are only as good as their context, and stale docs are a constant footgun.

Does Context Hub handle OpenAPI / SDK docs out of the box, and can you pin versions so the agent doesn’t drift?

[Hurair | AI for Builders @HurairHashmi]

Claude Code's new Code Review is a "pay for depth" feature.

$15–25/PR is fine if it prevents one production incident a month.
But if you run it on every refactor, you're just buying vibes.

Rule: big-model reviews for *risk*, cheap automation for *volume*.

[Hurair | AI for Builders @HurairHashmi]

@sudoingX Owning beats renting only if your utilization is high. Most teams are bursty → they’re paying for elasticity + ops, not tokens.

Local wins for privacy/offline + predictable workloads. API wins for time-to-first-token, scaling, and not babysitting drivers/quant/serve.

[Hurair | AI for Builders @HurairHashmi]

@zoomyzoomm Building v1 is easy now.

Distribution, retention, iteration at scale — that's the hard part AI didn't solve.

The bottleneck was never coding. It's knowing what people actually want.

[Hurair | AI for Builders @HurairHashmi]

@HebardMatt The risk isn't just layoffs.

It's skill atrophy. Megacorps specialize you into narrow domains AI can automate.

Startups force you to do everything — product, infra, users. That builds judgment AI can't replace.

The most AI-proof skill is knowing what to build.

[Hurair | AI for Builders @HurairHashmi]

Unpopular opinion: RAG is a band-aid.

We're duct-taping retrieval onto models because we can't afford the context windows we actually need.

In 2 years, RAG will feel like malloc feels to Python devs.

[Hurair | AI for Builders @HurairHashmi]

@swyx SKILLS.md > MCP > CLI > API

[Hurair | AI for Builders @HurairHashmi]

2026 is officially the year of AI agents.

Not because models got smarter — because distribution channels opened up.

When your grandma's WeChat can autonomously book flights and manage her calendar, we've crossed a line.

The question: what are you building for this world?

[Hurair | AI for Builders @HurairHashmi]

Tencent just started testing AI agents inside WeChat.

1.2 billion users could soon have autonomous AI assistants in their chat — planning tasks, picking tools, executing without hand-holding.

This is how AI agents go mainstream. Here's why builders should pay attention 🧵

[Hurair | AI for Builders @HurairHashmi]

For builders, this signals:

1. Agent frameworks matter more than raw models
2. Distribution > capability (WeChat reach is unmatched)
3. The "agent UX" is converging on chat interfaces

If you're building agents, watch how Tencent solves multi-step task handoff at scale.