Travers

we shipped the highest accuracy search API and the only truly built from the ground up for AI agents to use result is higher accuracy, lower cost, lower latency on end-to-end agent workflows we often talk about the AIs as our primary user, but what does that mean in practice?

@nickmartitsch @AnthropicAI yikes what’d you do

I love the Slack connector in Claude, but there REALLY needs to be a way to only enable read vs write cc @AnthropicAI

2026 is the year AI becomes truly stateful, persistent, and event-driven

@p0 this unlocks a ton

You can now create stateful web research agents with the Parallel Task API. Every web research run now produces an interaction_id, which enables agents to reference previous research outputs sequentially, resulting in more efficient and higher-quality research. Try interactions in the new developer platform showcase: platform.parallel.ai/play/interacti…

Anddd made my first transaction on @tempo! Searched using Parallel (@p0) to search top mentions of Artemis Analytics. It worked. Cost me $0.03. Set up took <1 min. Agentic payments is here and just beginning. Kudos @gakonst @matthuang and team ❤️

Parallel is built for a world where agents are the primary users of the web. Agents can now autonomously pay per API call for web access via Parallel using MPP. parallelmpp.dev

Agents will outnumber human users on the web by orders of magnitude. Just like people, they will need a way to pay for services they use. They may run into propriety health or finance data they need to pay for when doing a deep research task, or make a tool call to a bespoke web API for some functionality. But unlike people, agents experience no friction when making a payment, so they can pay for things in much smaller units and increments than people will. An agent may need to call an API that they only need to use on a one-time basis or pay for information that they need without signing up for a subscription. This means all forms of revenue streams can emerge for technology and information providers that wouldn’t have been possible before. To make this all work, we need will need new infra and tools for agents to do this, and it’s cool to see MPP from stripe and tempo.

@jeff_weinstein new ecosystem forming! has been so fun shaping this with tempo/stripe team

@travers00 so pleased to pair on this!

agents are the web's second user. parallel is infrastructure built for them to use it natively now they can pay for usage at runtime on the Machine Payments Protocol

Check out the interactive demo on the MPP docs site mpp dot dev

@pelaseyed @aisdk @p0 yup, this is the future. have been building similar things in different domains. this is super cool

Building out a CVE monitoring system with Infinite Monitor. Each widget it's own little agent environment powered by @aisdk @p0 just-bash and a bunch of other stuff I've been meaning to try out. Initially I wanted each widget to be sandbox, but it became far to expensive.

I think you are also right and true - my point was more observational of some of the key differences I’ve noticed and my own usage pattern in particular Opus is great at long running tasks reinforces the outer loop concept I’m talking about here - best for outer harness or complex reasoning impossible to get all the nuance in a single tweet

@travers00 I think the models have a much more interesting and high dimensional texture than this GPT 5.4 is an excellent writer, Opus 4.6 is incredibly reliable at long running tasks they are very different but in hundreds of more subtle ways imo and it can change between generations

i find it kind of interesting that anthropic and openai's models and distribution seem inverted openai models are more reliable/accurate, more consistent, better for structured data processing = "inner loop" work claude models are more expressive, open-ended, better at taking abstract human input and orchestrating the final output = "outer loop" work yet openai owns the consumer endpoint with 900m weekly users on chatgpt - that looks like outer loop, but most consumer usage is actually simple, directed tasks. anthropic wins on api adoption and enterprise spend, which is increasingly "inner loop" more and more inference is shifting to background sub-agents, reliable always-on processing - that's where the volume is heading, even if the margin is largest at the outer loop. and more and more of the inner loop may get consumed by open source inference anyway

@john_ssuh totally agree

@travers00 x.com/john_ssuh/stat…

this is why anthropic's microsoft-style shift makes sense forget the consumer. own the outer loop knowledge worker directly - the coder, the analyst, the powerpointer claude code, cowork, claude for excel, they're all starting to merge into a single interface for getting work done

@ankrgyl we have both and they work well in different contexts. our mcp is awesome in claude chat but cli is better in claude code or other harnesses conversational/open-ended vs agent-initiated and programmatic use

people who feel strongly on mcp vs. cli i am doing some research here. are there any particular services that have an mcp and cli (both) and you feel like one is way better than the other?

@khushi_shelat banger

idk about you but it all just CLIcked for me

created a pi extension for @p0 CLI - parallel.ai is really good/fast/cheap github.com/HazAT/pi-paral…

@DanielGri @p0 this is super cool, going to use this

Introducing the Parallel CLI: the most intuitive way for agents to access high-quality data from the open web. - Search: Search the web in natural language - Extract: Get clean page contents from any URL - Deep Research: Run deep research on open-ended questions - Enrichment: Enrich CSV or JSON data with AI-powered web research Read the blog and get started with a single command in your terminal: parallel.ai/blog/parallel-…

since we started, we've always designed with an agent as our primary user our new CLI gives an even more intuitive way for agents to use our web infra and do more real work w/ the open web - use it w/ your favorite harness