UnicornOne @unicornOneAI

@papercliping Thank you for the update

@dotta Nice update 👌🏻

The reason this is a hard question is it used to actually be either/or — both took months, so you were always starving one to feed the other. That’s the part that changed though. Product’s not the slow half anymore. So you lead with audience and let the product chase whatever they tell you they want. Audience first isn’t really a gamble once building is the fast part.

The model you build on can vanish for reasons that have nothing to do with you. This weekend that stopped being hypothetical. #Fable #Mythos It’s why we don’t bind our agents to a single model. They answer to a role, a goal, and a budget — the model underneath is a swappable part. One disappears, the work continues on the next. Model-agnostic by design. Built on Paperclip.

@dotta 100 PRs in a day is a wild stat for a project this young. From the building-on-top side it’s the whole case for betting on open source — the engine gets better every week whether we touched it or not. Half the features we leaned on this month didn’t exist in April.

Agree, but for a specific reason: what you actually learn at a startup isn’t your job. It’s how a company works — how tasks get scoped, handed off, reviewed, who owns what when things break. That knowledge matters more than ever now, because running a team is the founder skill. Doesn’t matter if the team is people or agents.

Company-shared skills is quietly the institutional knowledge problem getting solved. In a human org, what one person learns lives in their head and leaves when they do. Here one agent learns a workflow and the whole company just… has it now. Every new hire shows up pre-trained. That part’s underrated.

@danmartell yeah the memory part is the tell. a vending machine forgets you the second the can drops — every prompt starts from zero again. the system version actually holds onto context, keeps working when you’re not looking, owns stuff over time.

@bricktimesyt @VaibhavSisinty Oh c’mon

the fear makes sense if the job is “writing code.” it inverts if the job is “deciding what gets built and checking it.” your pricing point is the real headline — when frontier capability costs the same as last year’s, production stops being scarce. direction becomes the scarce thing. that’s a different job, not a missing one.

The “is it hype” question has a clean answer: it’s hype when the loop is the goal, real when the loop has an owner. A loop that exists because someone wired tools together = demo. a loop that exists because a specific job needs doing on a schedule, with a budget and someone checking output = just work, structured properly. The second kind was never hype. it just got cheap.

the one-shot demos are wild, but the quieter shift is what happens after the demo. a model this capable doesn’t need a better prompt — it needs a job. Paperclip shipped Fable 5 support this morning. same model, but with a role, a budget, and a ticket queue. that’s where this gets interesting.

New Paperclip release is a strong one. Fable 5 support, artifacts, and low-trust agents — that last one matters more than it looks. Scoped trust per agent means a prompt-injected worker can’t take the whole org with it. The engine under @unicornOneAI keeps getting better. nice work @dotta + contributors.

Paperclip @papercliping

a month ago

Paperclip v2026.609.0 dropped 💸 Anthropic Fable 5 support 🖼️ Company Artifacts and Rich Attachments ☑️ Checkbox Confirmations 🛡️ Low-trust agents github.com/paperclipai/pa…

2 2 24 12K 9

this is quietly a big deal — it makes the org chart itself a security boundary. same way you wouldn’t give a new hire prod access on day one. the intern who reads the inbox can get fooled, but the blast radius stays at intern-level. scoped trust per role is exactly how real companies handle this.

mostly agree, especially the “don’t run code you can’t read” part. that’s the real risk. i’d reframe the loop debate though: the problem isn’t loops, it’s loops with no oversight built in. your step 2 — “ask it what it did, concisely” — is an oversight loop. you’re not arguing against loops, you’re arguing for ones with a human check that actually works. which is the right argument.

Bring your own keys. We didn’t want to lock anyone into a single provider, so we built it the other way round — you bring your keys, you swap models whenever you like, and your company keeps running. We think your infrastructure should answer to you, not to us. Built on Paperclip. Try it: unicorn.one #Paperclip #buildinpublic

the reason this pairing is a big deal and not just two tools stacked: Hermes gives an agent memory and capability, Paperclip gives it a role, a budget, and a paper trail. capability without structure is just sprawl. structure without capability is an empty org chart. together it’s an actual employee.

Julian Goldie SEO @JulianGoldieSEO

a month ago

Hermes + Paperclip Just Changed AI Agents Forever x.com/i/broadcasts/1…

1 0 3 449 5

@omarsar0 goals + verification is the right spine. the thing that sneaks up on you: once you’ve got a clear goal, a budget, a verification step, and a record of what ran — that’s not an automation anymore. that’s a job. the “automation” framing undersells what you’re actually building.

nothing comes close to that first payment, agreed. and the reason “start marketing now” is hard advice to follow is that building is fun and marketing isn’t. so it’s the thing that gets pushed to next week, forever. the founders who win are the ones who treat it as a job that has to get done, not a thing they’ll get to.