The Missing Layer in Agent Systems Is the Organisation Itself
Oriens7 · reddit · 2026-09-27
After six months building OSIO — an operating system where people and AI run a business together — the author shares architectural lessons. The premise: if AI can do real work, an owner should state outcomes ('follow up late-paying customers', 'order 500 more units', 'make sure I send Acme the revised proposal tomorrow') rather than prompt task by task.
The deeper he got, the less the agent itself seemed the hard part. A capable model can reason about most requests; the harder problem is everything that must remain true around the model: who is responsible, what the AI is authorised to do, which systems it can use, what needs approval, what happens if the conversation ends or the process crashes, what if another model picks up the work tomorrow, and how you know something actually happened rather than the agent claiming it did.
His strong architectural position: the organisation shouldn't live inside the agent — the agent should work inside the organisation. Models, providers and agents change; the organisation must persist above them with its customers, people, responsibilities, permissions, commitments, work in progress and history.
He separates capability from authority: Gmail access proves an agent can send email, not that it is authorised to send this email, for this organisation, on this work, under these circumstances — the same applies to payments, purchasing, hiring and contracts. Agent memory, too, shouldn't be the source of truth: conversational context, evidence and organisational state are different kinds of information, and important commitments shouldn't vanish with a session. The goal isn't maximum autonomy but useful responsibility.
More from AGI Musings
- Anthropomorphizing AI: why 'think' is acceptable — with caveats — analisereal · 2026-09-27
- Argument: there won't be one ASI, but many locked in survival competition — JOBhakdi · 2026-09-27
- Timnit Gebru: deep learning folks ignore neuroscience and state guesses as facts — mjdramstead · 2026-09-27
- ASI skeptic: progress carries risks, but ASI brings infinite costs and zero payoff — JOBhakdi · 2026-09-27
- Beff Jezos: Future Will Marvel That We Once Did Knowledge Work Without AI — beffjezos · 2026-09-27
- Does ASI Have a Will to Power? The Core AI Existential Risk Debate — JOBhakdi · 2026-09-27