<![CDATA[Hung Hoang's Substack]]>https://www.hunghoang.mehttps://substackcdn.com/image/fetch/$s_!TTaA!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8afe848b-eef4-4980-88d6-7f7413c57ed1_144x144.pngHung Hoang's Substackhttps://www.hunghoang.meSubstackWed, 15 Apr 2026 05:56:13 GMT<![CDATA[Agency mode]]>https://www.hunghoang.me/p/agency-modehttps://www.hunghoang.me/p/agency-modeMon, 16 Feb 2026 08:38:08 GMTMost startups don’t lose because they picked the wrong idea.

They lose because they ran in low-agency mode:

  • waiting for permission

  • waiting for perfect clarity

  • waiting for “the right owner”

  • waiting for the next meeting

The shift that changed everything for me is Agency Mode:

Assume you’re allowed.

Assume you’re responsible.

Assume speed is a feature.

What Agency Mode looks like in practice

1) Start with the outcome, not the process

If a customer issue appears, the goal isn’t “file a ticket.”

The goal is: customer is unblocked today.

2) Replace handoffs with loops

The best teams don’t “throw” work over walls.

They close the loop end-to-end: observe → decide → ship → verify.

3) Default to a Version 1

In startups, a good V1 beats a perfect plan.

V1 creates reality. Reality creates clarity.

4) You don’t need more people - you need more agency per person

One high-agency engineer with tight customer feedback can outperform a whole committee.

Why this matters even more now

AI is compressing the cost of execution.

So the bottleneck is shifting:

less “can we build it?”

more “do we have the agency to decide, ship, and learn fast?”

At Zalos.ai, we’re building computer agents for finance operations—because finance teams are buried under approvals, exceptions, and “someone needs to…” workflows.

But the meta-lesson is universal:

The companies that win will be the ones where agency is the default setting.

]]><![CDATA[Staying Close at Scale]]>https://www.hunghoang.me/p/staying-close-at-scalehttps://www.hunghoang.me/p/staying-close-at-scaleThu, 08 Jan 2026 04:48:58 GMTAsked a small customer question in a YC × Vercel Slack.

Team replied.
Support offered.
CEO noticed.

Even at Vercel’s scale, a Series F valuing the company at $9.3B, the founder/CEO Guillermo Rauch is still close to day-to-day customer questions.

That’s not accidental. That’s how great products stay great.

On a related note, YC companies aren’t “early users.” They’re the sharp edge of the future. Listening early is how platforms stay ahead.

We are hiring an AI product engineer at Zalos (YC F25). We’re building the future of finance. If this kind of customer-obsessed, high-ownership environment resonates, feel free to reach out.

Thanks for reading Hung Hoang's Substack! Things I’ve learned building a company

]]><![CDATA[Thêm 1 cái tát vào mặt của Google và tại sao startups can win?]]>https://www.hunghoang.me/p/them-1-cai-tat-vao-mat-cua-googlehttps://www.hunghoang.me/p/them-1-cai-tat-vao-mat-cua-googleMon, 05 Jan 2026 17:52:14 GMTCách đây vài ngày 1 Principal engineer* ở Google đăng tải dòng tweet làm dậy sóng với hơn 7tr view và hơn 3.5k shares với nội dung đại loại như sau: chúng tôi mất gần 1 năm mà làm chưa xong mấy con agents mà chúng tôi muốn build. Bọn tôi loay hoay mãi với một vài lựa chọn...Tôi rảnh quá nên ngồi đưa cái đề bài cho Claude Code, nó mất 1 giờ đồng hồ để làm xong những gì chúng tôi mất hẳn 1 năm.

Thế rồi PG, founder của YC share lại cái tweet này bảo là, trong lúc mấy ông lớn ngồi chờ họp hành liên miên xong mà còn chưa làm được gì thì AI làm xong mịa cái version 1 mất rồi. Mà có lựa chọn nào khác ngoài cái ver 1 đó đâu. AI có rảnh đâu mà chờ.

Thanks for reading Hung Hoang's Substack! Subscribe for free to receive new posts and support my work.

This is a reminder that Startups can win. By what? Speed of execution.

*Cho những bạn nào chưa biết thì Princial engineer ở Google tương ứng với L8, trên Senior Staff và chỉ thua L9 là level cuối của Google.

Thanks for reading Hung Hoang's Substack! Subscribe for free to receive new posts and support my work.

]]>