We released Sonic-3.5 and Ink-2, the #1 streaming models for text to speech and speech to text you can use in your voice agents today.
New architectures enable new frontiers for speed and quality.
We're now the only provider to have #1 models for both speaking and listening.
Retool's new report says 35% of enterprises have already replaced at least one SaaS tool with something custom-built. 78% plan to build more this year.
I think about this number a lot. The conventional wisdom for the last 15 years has been "don't build what you can buy." And that was correct when building custom software meant hiring a dev team, spending months on requirements, and maintaining the thing forever.
But the calculus is genuinely changing. When a non-technical ops person can use an AI coding tool to build a working internal dashboard in an afternoon, the old framework breaks down. The build cost dropped by 10x, but the maintenance cost didn't. That's the part people aren't talking about enough.
The 35% who already replaced a SaaS tool are going to learn something interesting over the next 12 months: building is now cheap, but maintaining, securing, and scaling custom tools still takes real work. The companies that figure out which things to build and which to buy will have a meaningful advantage.
I don't think SaaS is dying. I think the bar for what justifies a SaaS subscription just got a lot higher.
https://t.co/rkPzN0kYTi
it’s very interesting to see consumer features get copied the moment they’re fun and obvious.
cursor building plan mode, codex featuring skills - good ideas converge quickly.
feels very uber/lyft - tipping, shared rides, etc - the other one picks it right after
note sure if I've seen anything promising so far. layered cryptographic proofs might work well for discrete info (expiry, spend limit) etc but get tricky when policies are more nuanced ofc. i think a big part of proving runtime integrity is also model / model weights. hopefully a standard gets defined and accepted else networks will have a frenzy
I made a framework for writing AI Agents as React components!
- Define agents declaratively with <Message>, <Tools>, nested agents...
- State works like you'd expect. Mount/unmount tools/messages based on conditions.
- <Condition when="<natural language>" /> allows changing behavior based on the conversation history.
Built on @AnthropicAI's API with support for code execution, web search, and MCP servers.
modern card systems rely on predictable human behavioral assumptions (aka vibes) to detect fraud.
e.g time of day → you usually spend between 9 am - 9 pm so a random charge at 3:47 am gets flagged
agents will be doing transactions at odd hours everyday.
at some point ‘weird hour’ will stop being a signal. agents will have .. different vibes
@t_xu said something I’m still chewing on - people who care deeply about craft also care deeply about efficiency.
That clicked for me because craft isn’t just taste it’s making good decisions under constraints around time energy attention reliability
In the Valley, people kind of already live this
more Whoops, more non alcoholic beer, more sleep and training treated like serious inputs not as a vibe but because it affects output
But until recently the bottleneck was still production capacity, team size, bandwidth
Now tools like Claude Code make execution dramatically cheaper so the constraint shifts from building to choosing what to build, why it matters, what quality bar to hold
And if the hard part is direction and clarity then personal efficiency becomes leverage not in a biohacking way just in a “protect your best thinking” way
Once those pieces exist, I think what'll be really interesting is how agent identity is modeled. I think the more easy-to-understand model is agents as scoped sub-identities of businesses / people (like highly programmable virtual accounts). I think the more exciting model is agents as independent counterparties with their own limits and reputations that networks can underwrite directly. The shape of that abstraction will decide how trust, risk, and fees get priced in an agentic economy
I think this is also why agent identity feels like two different problems: first, proving an agent is genuinely acting on behalf of a specific human or business (delegation, auth, cryptographic credentials); second, proving this is the same agent-policy bundle you interacted with before, not some sort of a clone or compromised instance
Spent some time understanding UCP yesterday. IMO it changes where the leverage sits vs ACP. A few margin notes:
Discoverability > ACP is more "app store": you list products via Stripe/OpenAI to show up in ChatGPT. UCP feels more like DNS for commerce: you publish /.well-known/ucp on your own domain and any compatible agent can find you. That shifts discovery power from platforms back to merchants, reducing long-term dependency risk - if one agent/platform changes their rules or cuts you off, you still exist for others.
Commerce scope > ACP is a traffic / checkout play: “I found the item, let’s pay.” UCP seems like a bolder bet on the messy rest of the journey - order tracking, returns, support via agents. Being able to ask your agent to “where’s my order?”, “start a return”, or “apply my loyalty discounts” against the same protocol feels like it changes retention, not just conversion.
Payment optionality > ACP effectively routes everything through a single stack (Stripe + ChatGPT) for product loading and payment flow. UCP is designed for choice: multiple payment handlers, multiple processors, and negotiation between agent and merchant about how to pay. That’s more work upfront but gives larger merchants and platforms more room to keep their existing rails and economics
Extensibility > ACP evolves like a product surface owned by a few players: new behaviors mean new spec + rollout. UCP evolves more like an internet protocol: platforms and merchants can publish new capabilities on top of the core spec and agents discover them, instead of waiting for the base to change
Netting it out: ACP is probably what most merchants will favor in the near term; especially if they’re already on Stripe and just want ChatGPT demand, it’s closer to “drop‑in and go.” UCP is easy to expose (publish /.well-known/ucp), but meaningful coverage requires clean product data, stable APIs, and keeping your site agent‑friendly, so the upside feels bigger but the payoff is slower.