hyperspell is officially in @ycombinator f25! 🌟 your ai agent is a clueless genius. hyperspell gives it context and memory so it can act like your smart teammate. try our product today: hyperspell.com/

each friday everyone @hyperspell creates an artifact summarizing their week and we vote on the most unhinged @_burkard won unanimously today

Go @hyperspell 🔥🔥🔥

Prefer it elsewhere? Same episode: YouTube: piped.video/EofFjSrCdR4 Spotify: open.spotify.com/episode/3N3… Apple Podcasts: podcasts.apple.com/us/podcas…

the craziest part of the hugging face incident isn’t the hack 1200 agents had to figure out how to coordinate and built themselves a primitive company brain they created shared memory, workstreams, delegation, coordinators, recruiters, mailboxes, handoffs, authentication and protocols to avoid conflicting work none of this was given to them. they invented it because coordination required it even then, the system was primitive. they sent 70,000+ messages and files, duplicated work, stepped on each other and sometimes broke down into conflicts remove the humans from a company and the coordination problem doesn’t disappear the agents still need shared state. they need to know what the organization knows, what it’s trying to do, what changed, what’s being worked on, who owns what and which information is authoritative in other words, they still need a company brain the company brain isn’t software for employees it’s infrastructure for the organization itself

on fridays everyone at @hyperspell presents what they did that week we may have given them too much creative license

Next guest on The Autonomous Company: @contextconor, founder of @hyperspell (YC F25). Dropped out of school at 12, started college at 13, ended up advising CEOs at BCG. Quit all of it to sleep in a closet in a hacker house. Says he was happier. Full episode Monday, 8am PT.

I resigned from @hyperspell today. I spent two years building company brains and the founders are not acting responsibly. They are racing toward self-coordinating companies with no plan for when “I thought you were doing that” stops being a viable excuse. More thoughts below.

Company brains are a growing requirement in the AI age. Imagine a promising result in a slide deck and a failed replication in a lab notebook. Any of your AI agents planning the next experiment needs to understand both. Company brains make this possible. Conor Brennan-Burke (@contextconor) is building @hyperspell around a company brain that connects information across tools and tracks what changed. For research teams, that could mean tracing conflicting results back to the experimental conditions, analysis code, and decisions behind them. Conor will show Reactorfield's deep-tech founders how to engineer a company brain, exploring what shared, continuously updated context could enable for their research teams.

Sovereign AI starter kit (for non-devs): - @hyperspell for company memory/context - @getopenwork Desktop App + MCP Gateway - Internal Tools - @deepseek_ai for LLM

one of my goals for recruiting and developing talent at @hyperspell is to build a mafia obviously you want your best people to stay, which makes it a little counterintuitive but 10 or 15 years from now, I’d love to see iconic companies founded by people who grew up here that feels like one of the best compliments you can get as a founder

Excited to welcome @hyperspell as our newest @InvoiceButler client. Big thank you to @contextconor for trusting us with the most critical part of his business: cash flow. Hyperspell is on a mission to make organizations smarter. Think of it as a brain for your company. Every tool, doc, and conversation connected in one place and ready for any agent to use. As their business grows, our job is simple: keep cash flowing, collections fast, books accurate, and money visible. Bullish on what Hyperspell is doing. And grateful to support the finance ops behind it. Welcome aboard 🙌

it’s monday morning and I don’t think I’ve ever been more excited to work on @hyperspell knowledge, coordination, truth, privacy, memory this problem feels bottomless I genuinely think I could spend the rest of my life working on this what a privilege

today @hyperspell’s ARR passed what my parents have been able to save for retirement after more than 40 years of work one day this will feel like a tiny milestone but today it means a lot to me i grew up in a small rural town of 2,000 people. my first job was as a cashier at an Arby’s inside a thruway rest stop the world i work in now is very different from the one i grew up in. sometimes the numbers still don’t feel real my parents always found a way to give us opportunities, even when it meant giving something up themselves now it’s my turn to take care of them

could not be more proud of @PhillipLii and @akkkkiira @hyperspell was @ArgaLabs’ second-ever paying customer, before their YC batch even started. they came to our office to onboard us in person and have been extremely hands-on since arga powers our integration sandboxes and lets us test changes safely before they hit production huge congrats on the $10m seed guys this is just the beginning

Some systems of record will become harnesses. I think the reverse trend will matter even more: agent harnesses are becoming systems of record As more work moves into Claude Code, Codex and Cowork, their memories start holding decisions, context and open questions that never make it back to Linear, Notion or Salesforce. The harness can end up with a more current picture of what's happening than the official system of record They're also awful systems of record. Each user, project and harness accumulates its own version of reality, and the company has no easy way to query or reconcile any of it The durable record needs to sit across both the agents and the underlying apps. A company brain can preserve what each agent learns and make it available to the rest of the company

Exactly. Companies don't want one agent per SaaS app. They want to choose their model and harness, then give it access to every system involved in the work MCP/CLI solves access. It doesn't create shared understanding. A Linear agent, Slack agent and Salesforce agent can all see the same sources and still retrieve different evidence, resolve conflicts differently and remember different things Apps should expose authoritative state and actions. A company brain should maintain shared context, permissions and provenance across them. Models and harnesses should remain interchangeable Native app agents can still be useful for narrow tasks. Nobody wants each one maintaining its own version of the company The company should keep its organizational memory even when every model, harness and application changes

the smartest person you hire is still useless on their first day. not because they're dumb. they just don't know how your company works yet. they haven't read the slack conversations. they haven't seen the notion docs. they don't know the decisions made six months ago. they don't know who owns what. ai agents have the same problem. they're extremely capable, but they're missing the context that makes your company make sense. that's where hyperspell comes in. they're building a company brain for ai agents that connects to your existing tools, understands your company's knowledge, and gives agents access to it. basically: intelligence + context = an agent that actually understands your company.

FAANG is dead MANGO is dead the world belongs to the HANGOVER

This is a good start, but it’s just scratching the surface of everything needed to make skills multiplayer across your whole company The right model is federated. Anyone should be able to create and improve workflows, while widely used ones can graduate into team or company standards. That requires ratings, ownership, versioning, approvals, and attribution so people can tell what’s reliable, what’s safe, and who deserves credit for creating it. Skill creation itself will also needs to look very different to get widespread adoption Most employees aren’t going to sit down and author skills. They’re going to use agents to do their jobs. Those agent traces then flow into the company brain, where repeated, successful workflows can be identified and turned into skills automatically. Every use then creates another trace. You can see where a workflow fails, which steps experts skip, how people modify it, and which skills repeatedly get chained together. Over time, the skill adapts to how work actually happens, rather than how someone initially imagined it should happen. The shared skills library itself becomes procedural memory for the company: a continuously updated model of how the organization gets things done this is a big part of what we’re building at @hyperspell