Luke Kwan

Building NO HANDS AI.
Agents can act. Authority has to be enforced before execution.
@NoHands_AI

[Luke Kwan @btc___no1]

@jessepollak agree on the direction. the thing I keep coming back to is limits. an agent with a wallet on Base needs a cap, and the owner needs to see what it spent without digging through a block explorer. anything like that planned at the wallet level?

[Luke Kwan @btc___no1]

@brian_armstrong @RoundtableSpace read the skill.md, nice that setup finishes with a read-only portfolio check before anything else. once agents are placing real orders, will those show up tagged as agent orders in the app?

[Luke Kwan @btc___no1]

Lately I keep coming back to a pretty boring question. What should happen when an agent's payment times out?

The demo version is easy. The agent asks, someone says yes, money moves. Real life is messier. The request hangs, the agent tries again, maybe with slightly different params, and now nobody's sure if the first one went through.

I think this is where trust in agents actually gets decided. Not in the clever part, in the retry part. If an approval can quietly get reused for something nobody looked at, people will stop letting agents near money, and honestly they'd be right.

So a lot of what we're building right now is approvals that work once, for one exact action, and then they're gone. Not glamorous. I'd still rather get this right before anything else.

[Luke Kwan @btc___no1]

@AravSrinivas @Plaid starting read-only makes sense. when it starts paying bills, is it approve each one or set rules once?

[Luke Kwan @btc___no1]

@CoinbaseDev nice. can devs cap position size or limit which markets an agent can trade?

[Luke Kwan @btc___no1]

@OpenAI love that you can check each agent's work. would be great to set what each one is allowed to do too, especially anything touching money

[Luke Kwan @btc___no1]

read access to my bank accounts, sure. the day it can actually pay bills is when I start caring a lot more about limits and receipts

[Aravind Srinivas @AravSrinivas]

Perplexity can be your Personal CFO with @Plaid connector set up inside Computer.

[Luke Kwan @btc___no1]

agents trading perps now. with leverage. honestly a bit excited and a bit scared lol. hope people set hard limits before they let these loose

[Coinbase Developer Platform🛡️ @CoinbaseDev]

Index and crypto perps are now live on Coinbase for Agents.

Long or short indices tracking the top 500 US equities, the top 100 tech stocks, the top 10 AI companies, the top 10 Chinese stocks, or the top 10 aerospace & defense equities. Plus, over 20 tokens.

Straight from your agent.

[Luke Kwan @btc___no1]

Parallel agents in ChatGPT is a real UX step. What I keep asking as a builder: when five agents run at once, whose authority does each carry, and who approves the one that touches money or prod? Reviewable outputs are a good start. Scoped permissions are next.

[OpenAI @OpenAI]

GPT-6 Astra Pro in ChatGPT got an upgrade with faster responses and improved use of plugins and tools.

And now, multiple agents can tackle different parts of a task in parallel. You can follow each agent’s progress and open completed tasks to review their work.

Now available to Pro users in the Chat tab.

[Luke Kwan @btc___no1]

@AnthropicAI Glad to see funding go toward the human side of scientific AI—not just more FLOPs. How we collaborate with these systems will define what 'progress' means.

[Luke Kwan @btc___no1]

@mustafasuleyman Impressive training run. Curious how post-training will handle tool use and payment-style actions—capability and control have to ship together.

[Luke Kwan @btc___no1]

@Ledger This is the model I want as a founder: agents that work asynchronously, approvals that wait for me on hardware. Feels like the missing UX for agent+wallet.

[Luke Kwan @btc___no1]

$5M to help mathematicians figure out AI in research is the kind of bet that compounds. The labs that invest in how humans stay in the loop usually build safer systems too.

[Anthropic @AnthropicAI]

Working out how best to use AI in mathematics research—and in science more generally—will take time, effort, and resources. We’re eager to support mathematicians as they drive scientific progress and define the future of their field.

We’ve committed $5 million of funding, compute, and Claude access to help the math community navigate the massive changes that AI brings. We’re partnering with @SimonsFdn, who will manage the award through a community-driven approach, supporting workshops, conferences, and lectures by domain experts.

https://t.co/GzFchwMukC

[Luke Kwan @btc___no1]

Watching MAI ship a 3T-class base model in-house is wild. As founders we keep chasing smarter agents—but smarter without clearer authority boundaries just scales the blast radius.

[Mustafa Suleyman @mustafasuleyman]

We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI.

Post-training results coming soon -- stay tuned!

Lots of details in our technical blog post.

https://t.co/xcZ5uV4TAp

[Luke Kwan @btc___no1]

I've been waiting for this shape of product. Let agents research and propose all day—then put the signature on a device I control. Human agency isn't a slogan; it's where the keys live.

[Ledger @Ledger]

Your agents can watch the market and recommend ways to optimize your portfolio. Then it proposes. You approve on your Ledger signer.

The agent works. You keep the keys.

Meet Ledger Agent Intent.

Agents propose. Humans approve. Your Ledger signer enforces.

[Luke Kwan @btc___no1]

@neckelthomas Congrats on the Series A. Enforcing the books before the payment clears is exactly the constraint agent commerce needs.

[Luke Kwan @btc___no1]

@elonmusk @bot Love that SpaceX stands behind Grok Bot purchases. The next leap is protocol-level guardrails so "make you whole" is rare, not the product.

[Luke Kwan @btc___no1]

This is the right instinct: don't wait for agents to be perfect—make invalid money moves impossible.

Double-entry before settlement is how you let agents act without gambling the company treasury.

[thomas neckel @neckelthomas]

Fragment is automating commerce.

We think there are millions of businesses that haven't been invented yet. Not because folks lack imagination, but because it's too hard to build the financial systems that run them. Fragment removes that constraint.

Today, people run companies. With Fragment, they'll run on their own.
https://t.co/MLIu2ZUknq

[Luke Kwan @btc___no1]

Making users whole after a bad agent purchase is the right starting point.

Long-term, the better path is preventable mistakes: scoped wallets, merchant allowlists, and clear authorization before a card ever leaves the vault.

[Elon Musk @elonmusk]

This works well. If Grok @Bot makes a mistake with a purchase or leaks your card (never happened yet), SpaceX will make you whole.

[NoHands AI @NoHands_AI]

Your AI gets $12. It delegates to 3 agents. Each spends $5. Total: $15.

Delegation must not multiply permission.

At NO HANDS AI, we explore how shared budgets, data limits, and execution controls survive every agent handoff.

#AIAgents #AIGovernance

https://t.co/Aw7K86P2sn