Jason Han @JasonHan001

Dubai, United Arab Emirates Joined June 2025

Tweets

16
Followers

1
Following

72
Likes

0

Jason Han @JasonHan001

3 days ago

One thing should stay centralized: who gets to declare something complete. An agent can say “I’m done.” That means candidate complete, not system complete. I’ve seen Tasks advance while Program state stayed stale, and CI go green for a different revision.  So I’m cautious about one word: DONE. Doing the work and proving it is complete are different things. The same applies to health: Heartbeat ≠ Progress. An agent can be online and burning tokens while artifacts, evidence, blockers, and user outcomes don’t move.  There’s also over-governance. One mistake → Reviewer. Another → Controller. Then Audit, Validation, Approval. Soon the workers get stronger while the process gets longer. Governance should be proportional to risk. A local change shouldn’t trigger full-system revalidation.  Recovery shouldn’t mean “rerun everything.” Find what actually changed. Find the first unproven edge. Resume from there. My question changed from “How do I get more agents to work together?” to “How do I stop managing every agent while the system still delivers?” The future isn’t more AI employees. It’s an AI organization that can actually deliver.

0 0 0 4 0

View Details

Jason Han @JasonHan001

3 days ago

Vibe coding isn’t about playing boss to a bunch of agents. If you still assign every task, resolve every conflict, notice every stall, and verify every “DONE,” you haven’t built an autonomous team. You’ve just created more AI to manage.  What should transfer from human organizations into agentic development isn’t job titles. It’s management capability: clear goals, authority, dependencies, handoffs, acceptance, and exception handling. That’s what makes an organization actually work.  So I now reverse the usual order: Don’t design a fixed team of agents first. Define the delivery outcome first. Then assemble the capabilities needed to reach it. Use one agent if one is enough. Parallelize when useful. Add a reviewer only when needed.  My view of the central controller changed too. It shouldn’t be a super-agent that micromanages implementation. Its job is to control WHAT, constraints, priorities, and acceptance. The HOW should stay with execution agents as much as possible.

1 0 0 6 0

View Details

Jason Han @JasonHan001

4 days ago

@Careerbuddha111 I’m also a developer too. Glad to connect.

0 0 0 13 0

View Details

Jason Han @JasonHan001

5 days ago

@typesafeai Faster and cheaper is real progress. But should “next-gen AI” be defined by removing humans from the loop — or by improving the decisions humans and machines can actually make?

0 0 0 292 0

View Details

Jason Han @JasonHan001

5 days ago

AI is most valuable to me when it changes the decision space itself: a missing option, an untested assumption, new evidence, or the realization that we shouldn’t decide yet. That matters more than simply reaching the old answer faster.

0 0 0 3 0

View Details

Jason Han @JasonHan001

5 days ago

So my test for the next generation of AI is simple: Under comparable information, budget and risk, does it improve the decision — or deliver the same decision quality at lower total cost? Verification, human review, rework and mistakes all count.

1 0 0 4 0

View Details

Jason Han @JasonHan001

5 days ago

Jev is faster. That doesn’t make it the next generation of AI. The real test isn’t tokens/sec or getting humans out of the loop. It’s whether AI changes the decision you were capable of making — by revealing options, assumptions or evidence you couldn’t see before.

TypeSafe AI @typesafeai

5 days ago

Jev is now available to everyone. No waitlist. Start using it here: console.typesafe.ai

729 2K 23K 3.1M 12K

1 0 0 55 0

View Details

TypeSafe AI @typesafeai

5 days ago

Jev is now available to everyone. No waitlist. Start using it here: console.typesafe.ai

729 2K 23K 3.1M 12K

View Details

Jason Han @JasonHan001

2 weeks ago

My products are still unproven. So this is not a victory post. But it is a milestone for me. For the first time, I can say with evidence: I’m not just using AI tools. I’m learning how to orchestrate AI-native engineering. And yes — I’m genuinely proud of these two months.

0 0 0 1 0

View Details

Jason Han @JasonHan001

2 weeks ago

That was the shocking part. Many of these governance practices were lessons I learned by troubleshooting real project issues: wrong versions, fake “done,” repeated loops, broken handoffs, missing persistence. Today, I am aligning with frontier-style agent engineering.

1 0 0 1 0

View Details

Jason Han @JasonHan001

2 weeks ago

I’m not a software engineer. My background is finance and audit. But after ~2 months of heavy Codex development, I finally looked at my actual GitHub data — and it genuinely shocked me: 44 days 244 PRs opened 206 merged 5.55 PRs/day 84.4% merged