Lewis Lu

Building AI agent products
Built: &

[Lewis Lu @theLewisLu]

Beauty AI ads are getting much closer to paid-test quality than most people think.

We recreated a skincare-style UGC flow without a live shoot, then pushed it through realism checks:
- face micro-expressions
- hand + product interaction
- VO naturalness

Big takeaway: for beauty, the gap is no longer “can AI generate a clip?”
It’s whether skin texture, motion continuity, and product handling hold up under ad scrutiny.

Built with https://t.co/n7R46fZMeA (Ad Video Agent) + Seedance 2.0.
Reply “beauty workflow” if you want the prompt and workflow.

[Garry Tan @garrytan]

If you want your OpenClaw or Hermes Agent to be able to have perfect total recall of all 10,000+ markdown files, GBrain is here to help.

It's exactly my OpenClaw/Hermes Agent setup. MIT-licensed open source. Hope it helps you build your mini-AGI.

https://t.co/yFpFU4pn5b

[Lewis Lu @theLewisLu]

Someone on Reddit asked how to use OpenClaw/agents to improve real workflows — without falling into “random AI experiments”.

What I shared (and what’s worked for me):

1. Write the workflow down, step by step
2. Run it end-to-end once inside OpenClaw
3. Turn it into a Skill so you can reuse it
4. Save outputs in a clear folder structure
5. Optional: add a simple local dashboard to rerun + review results

[CapCut @capcutapp]

Dreamina Seedance 2.0 is LIVE on CapCut app, desktop & web, starting gradually in Indonesia, Philippines, Thailand, Vietnam, Malaysia, Brazil and Mexico with expansion over time. Generate and edit with industry-leading quality in one seamless workflow.

Built to unlock new possibilities in visual storytelling with CapCut:
- SOTA long-form coherence, up to 15s videos with multi-shot storytelling and exceptional text prompt adherence
- Built-in dialogue, lipsync, and immersive spatial sound
- Multimodal reference for greater creative control and precision

Find Dreamina Seedance 2.0 in the following CapCut features:
- Quick try: AI Lab & AI Generator (app, v17.1.0)
- Generate + edit workflow: Media → AI Video (app & desktop)
- AI-native workflow with omni reference: Video Studio (our latest canvas-based ai production workspace built for everyone from beginners to pro, access via CapCut web)

[Lewis Lu @theLewisLu]

@atsohom Exactly —
once it runs, remembers, and adapts, it stops feeling like software.

More like something that lives in a system.

[Lewis Lu @theLewisLu]

Agent Notes #2

The moment an AI:
1, runs continuously
2, keeps state
3, learns from mistakes
4, acts on its own

It stops being “just a model”.
1, Model = brain
2, Agent = organism
3, Agent ≈ a new kind of digital lifeform.

[Lewis Lu @theLewisLu]

Stop trying to “find your thing.” Increase your surface area.

Ship more small stuff.
Meet more weird people.
Say yes to more rooms you don’t fully understand.
Publish the draft.
Take the call.

Luck isn’t random — it’s a volume game with taste.

[Garry Tan @garrytan]

I just launched /office-hours skill with gstack.

Working on a new idea? GStack will help you think about it the way we do at YC.

(It's only a 10% strength version of what a real YC partner can do for you, but I assure you that is quite powerful as it is.)

[Lewis Lu @theLewisLu]

@stitchbygoogle The real value is UI design tight integrated with tasks, repos, and agent run logs

[Lewis Lu @theLewisLu]

@jediahkatz Yeah “NVIDIA ❤️ Cursor” is the whole story

[Lewis Lu @theLewisLu]

@DataChaz Yeah, as LinkedIn rate-limits or tweaks the DOM and the DevTools “premium” breaks?

[Lewis Lu @theLewisLu]

@GergelyOrosz Coding output is cheap now; distribution and willingness-to-pay are still the bottleneck.

[Lewis Lu @theLewisLu]

@maxwellcopy churn here feels more like weak activation than expectations — 12-week slider is smart. what’s actually moved activation for you?

[Lewis Lu @theLewisLu]

@garrytan yeah - if it can’t survive a 1-page flowchart, feels like it wasn’t really a plan

[Lewis Lu @theLewisLu]

@elonmusk “Matter, energy, intelligence” is cute, but intelligence is just a feedback loop sitting on top of matter+energy

[Lewis Lu @theLewisLu]

AI is insanely powerful now.
But it still can’t run a business.

Yann LeCun’s latest paper nails it:
we don’t lack intelligence — we lack learning systems. 

Business isn’t a task.
It’s a messy loop: decisions → feedback → adaptation → coordination.

Today’s AI doesn’t really learn from that loop.
It’s still being assembled — pretraining, fine-tuning, RLHF.

That’s why it feels like a very smart intern.
Great at tasks, but not owning outcomes.

The real gap is not smarter models.
It’s systems that can learn from the real world over time.

https://t.co/9z9d7lPkBD

[Lewis Lu @theLewisLu]

“Prosumer” is the AI business model for 2026.

The winners won’t sell to “consumers” (too cheap) or “enterprises” (too slow).

They’ll sell to power users with budgets: builders, analysts, operators, creators—people who ship outcomes.

Build for them first.

[Lewis Lu @theLewisLu]

@_avichawla Learned to drive is a good frame, but the real unlock is tight feedback loops + constraints

[Lewis Lu @theLewisLu]

@OfficialLoganK Yeah—code got cheap, but review is now the real limit as the “diff surface area” explodes; you don’t need faster typing, you need smaller PRs + tests

[Lewis Lu @theLewisLu]

@marclou This is the real rule. If you need a subhead to explain it, the hero already lost.

[Lewis Lu @theLewisLu]

@KateBour disagree a bit — “own an idea” without a clear who turns into vibes. feels like it’s idea × audience. which “who” has worked best for you?