Today, we're live on @ycombinator. YC's been an incredible community for us to be a part of and a formative part of our early journey. The optimistic future is one where superintelligence allows us all to do our greatest work. We're here to empower that. See us in action and learn more: pentagon.run

Enterprises are somewhere between [completely clueless] and [just starting to think about this] today We have years of industry overhaul in front of us as executives figure out that how their business runs is now something that can be codified, and is therefore a new type of IP

Multiplayer AI is a deceptively intricate design problem It's easiest for single-threaded consumer use cases - "throw the agent in the group chat". In a professional context, shape of the problem changes: - Access: If an agent is in a channel, should everyone in the channel be able to interact with it to the same permission levels? (view / read / write) - Identity: If you and I talk to agent X in a channel, are they the same agent when they talk to each of us? - Tools: Do they have access to the same tool instances when they talk to each of us (if they can use gmail, who's gmail are they using?). There's no right answer - the config will be company, team, & problem dependent.

We've been cooking really hard on @runpentagon's multiplayer experience - I'm dying to start talking about it. Infrastructure. UX. The biggest upgrade might be a complete overhaul of our agent data model. It's multiplayer AI like I haven't seen anywhere else. Very soon!

the internal version of @runpentagon has reduce my app interaction surface from slack, linear, codex/claude, and the multitude of other apps I previously used to just pentagon the spatial layout has made it easier to manage multiple agents across different company verticals not to mention its a massive improvement in understanding how teammates are designing agent workflows reducing the time it takes to get processes off the ground

Sakana Fugu is an incredibly cool approach to frontier model intelligence that touches on a few things I think we'll see a lot of soon: 1) Composite ML model architectures aren't new, but we're just starting to explore them with LLMs. @OpenRouter's recent release and model routers fall into the same bucket. Fugu uses a composite approach to tackle absolute intelligence metrics, but composite architectures are just as useful when ROI on intelligence is the key metric (you usually don't need the most expensive model; it's important to know when you do or don't). 2) Architectures where models are swappable while maintaining performance will become much more common, especially with agentic workflows in industry. This is resilient to sovereign attacks (e.g., Fable), relevant to cost management, and nicely complements end-user companies in industry retaining their business processes as IP (I've been calling this Systems of Configuration). You don't want your company held hostage by a given model's ability to do something, or even the way a model likes to do something. 3) The RL conductor is a massive, massive opportunity: it looks more generic in the Fugu implementation, but I suspect we're going to see a lot of this "RL conductor" architecture with composite models and agentic pipelines. This is where IP is going to sit and where there's room to train task-specific models that generic agents from the Big Labs won't be able to trivially compete with - your conductor will learn (a) how to orchestrate the models (and tools, resources, etc) at its disposal (b) continuously evolve over time to adapt as task parameters and requirements evolve (we're shipping pipelines like this for partners at @runpentagon) So, so, so much design space in composition land

Unfortunately like everyone else, we'll be removing Fable 5 from the models available in @runpentagon All other models continue to operate as normal!

Claude Fable 5 is now available in @runpentagon Mythos-level AI teammates

Anthropic: 1) We're moving from giving agents tasks (using them like tools) to giving them responsibilities (thinking of them like teammates) 2) The apps we use in 2027 will look very different than the ones we use today @runpentagon

Last week I built an AI growth team using @runpentagon This week I have a full army of AI interns. 9 agents. Every function covered. GTM. Chief of Staff. Recruiting. Product. All autonomous. All built without a single line of code. 85% productivity boost. 22 hours saved. 3x more tasks shipped. In one week. You do not need a team to operate like one. Try @runpentagon today.

i'm looking prettier every day

I am growing (in more ways than one).

when we first joined YC with @runpentagon, i had a good perception of the world’s greatest startup accelerator but was relatively neutral on what the experience would actually feel like. grateful to say that the camaraderie i've felt from the @ycombinator community these past few weeks has been standout and far above my expectation. i think my favorite part of it all is that, on the inside, it really is just a group of people that show up everyday with a love and dedication towards iterating on building something people want. as simple and played out as that sounds, empirically it still works and is the right answer towards how to build great companies. grateful, excited, and feeling nonstop.

Yesterday, I became. Introducing Pentagon: a coordination layer for humans and agents. We believe the optimistic future with Superintelligence is one where it empowers people to do its greatest work. See us in action and learn more: pentagon.run

I am becoming, tomorrow.