Your Agent Can Run Billing Now. It Still Can't Hit Send.

Quidbill started with a narrow promise: create a professional invoice in about 30 seconds, without fighting an accounting suite to do it. That promise still stands. This post is about what we built on top of it, and why.

The bottleneck moved

If you run a solo business in 2026, there is a decent chance an AI agent already sits somewhere in your stack. It reads your project notes, drafts your emails, updates your task board. The hours you used to lose to admin are increasingly hours your tools spend instead.

Billing has resisted that shift, and for a good reason. An invoice is not an internal document. It goes to a client, with your name on it, asking for money. A typo in a task board costs you a minute. A wrong invoice costs you trust, and sometimes the client.

So most tools picked one of two answers. Either agents get no access to billing at all, or automation sends things on your behalf and you find out afterwards. The first answer throws away help you already have. The second one is how you end up apologising to a client for an invoice you never saw.

We think both answers are wrong, and the right one is structural: give the agent real access to the work, and give the human a gate the agent cannot cross.

What Quidbill is now

Quidbill is billing software your agent can operate. Concretely:

• An Agent API. A bearer-authenticated HTTP surface with an OpenAPI contract. Your scripts and agents can list clients, list invoices, inspect an invoice with its line items, and create draft invoices from real client data.
• An MCP server. One command (bunx quidbill-mcp) connects Claude Desktop or any MCP client to the same four tools. Your agent works from the records you already keep in Quidbill, instead of hallucinating client details.
• A human approval gate, enforced by the product. Every agent-created draft stops in a review queue. You see exactly what the agent prepared. Approval changes the invoice state; sending remains a separate action that belongs to you. This is not a prompt-engineering convention. The API has no way to skip it.

The manual desk did not go away. It remains a focused workflow, and the product works fine if no agent ever touches it. The agent layer is optional, never a requirement.

Why the approval line is the product

The obvious question: if the agent can draft, why not let it send?

Because drafting and sending are different kinds of actions. A draft is cheap and reversible. You can discard it, edit it, or laugh at it. A sent invoice is outward-facing and irreversible in the way that matters: your client saw it. The cost asymmetry is enormous, so the control should sit exactly on that boundary, and nowhere else.

Putting the gate anywhere earlier throttles the useful work. Putting it anywhere later is theatre. One line, drawn once, enforced by the product: agents prepare, humans approve, then it can be sent.

This is also why we say the approval line is the product. Plenty of software can render an invoice PDF. The thing a solo founder actually needs in 2026 is a way to accept agent leverage in the one workflow where a mistake reaches a paying client.

What this is not

Scope honesty, so you can qualify us out quickly:

• Quidbill is not an accounting suite. No bookkeeping, no reconciliation, no payroll. If you need those, use a suite alongside it.
• Quidbill does not collect card payments from your clients. You set the payment instructions that fit each client, and track payment status in the desk.
• Quidbill does not auto-send anything an agent created. That is the whole point.

Where this goes

The direction is simple to state: every part of billing prep that is safe to delegate should be delegatable, and the parts that are not should be visibly, structurally human. Client records, drafts, reminders and payment status already live on the agent surface or next to it. The approval line stays where it is.

If you want to try the workflow: the free tier includes three manual invoices with no card required. Pro is $29/month for the manual desk. Agent is $49/month and adds the Agent API and MCP access. The Founding plan holds Agent access at $449/year for the first 25 active subscriptions.

Your agent drafts. Nothing leaves without you.