Why now

The shift

For two years we called them copilots. The framing was honest: in 2024, large language models were assistive. You typed; they suggested. You decided.

That framing is breaking. Tool use, in 2025–2026 frontier models, is reliable enough that the model itself can take the wheel — write code, send mail, charge a card, file a return, follow up next week. The pilot has stepped out of the cockpit.

What's left is no longer a copilot. It is an agent.

The wedge

But "agent" by itself is not a market. Vendors are racing to ship agents that do exactly what their copilots already did, and charging on usage. That is the SaaS playbook with one extra word.

The wedge is ownership.

A copilot belongs to the vendor. An agent — once it has memory, a wallet, a cap table — has to belong to someone. The runtime that lets you own one outcompetes the runtime that lets you rent one, because the equity story is structurally different. Renters scale linearly with usage. Owners compound.

Why now

Three things landed in the last eighteen months.

Models reliable enough at tool use to step out of the cockpit.

Sandboxing mature enough to let an agent hold a shell without holding the keys.

On-chain primitives — wallets, cap tables, vesting — programmable enough to back a real company.

None of these are individually new. The combination is.

Four primitives

aeqi runs on four nouns.

• Agents — persistent identities in a tree.
• Ideas — a unified store of instructions, expertise, memory.
• Quests — units of work with state, dependencies, outcomes.
• Events — when an agent acts.

Together they describe an autonomous unit small enough to launch in one command, big enough to scale into a company.

The closing claim

One person, a tree of owned agents, and an open runtime — that is the smallest viable autonomous company.

The runtime is the missing piece. We are building it in the open.