One-Person Startup: How One Founder Runs a Team of AI Agents

A one-person startup is one founder covering the functions a small team would split: customers, support, the product, the books, and the week itself. On 23 October 2026, Greg Isenberg posted a list titled "The MOST obvious (but REALLY good) STARTUP ideas right now". Idea #3 is the line this page takes up.

Y Combinator for 1 person companies (back solo founders running a team of agents)

Greg Isenberg, 23 October 2026. Idea #3.

xSquads is the operating side of that line. It gives the solo founder the team of agents. You import a squad for the job in front of you, point it at a folder you already have, and approve what is allowed to leave. xSquads does not invest, does not run a batch, and does not take a stake. You run the squads.

Short answer

Give every knowledge worker their own xSquad. In a one-person startup the knowledge worker is the founder, and the squad changes with the hat. xSquads for Work is the desktop app. The squad works on your own files and folders. Cards move from Backlog to To Do to In Progress to Review to Done. Bots propose, you approve. Nothing is sent, published, dialled, or merged until you approve it. Mac now, Windows waitlist open.

A shortlist of products for a founder, including tools that are not this one, is best AI agents for small business and solo founders. This page is the operating shape: several squads, one person, a Review column on every send.

What a one-person company actually needs

Headcount of one does not delete the functions. It stacks them on the same person.

Customers still have to be found, and the outreach still has to be written. People who already bought still write in. The product still changes. The numbers still have to line up. The same person publishes, chases the vendor, and clears the inbox before the only meeting of the day.

A chat window can draft any one of those. You then copy the answer out by hand. A one person company needs the draft to land in the folder, with a named role on it, and a column where it waits. On this product that column is Review.

How squads map to a one-person startup

Thirteen templates ship on squads. The seven below are the functions this page maps. A template is a starting point. Keep the agents the mission needs, and remove the rest.

The worked tutorials are how to run a GTM squad, how to run a support squad, how to run a content marketing squad, how to run a weekly reporting squad, and how to run an inbox squad. Marketing also has tutorials for SEO, social, and ad creative. Operations also has vendor follow-up. The chief of staff page links the weekly update and meeting prep. Software engineering and finance have squad pages, and no separate how-to-run tutorial on this site.

A few names, so the table is not only a headcount. On go-to-market, Scout builds the prospect list, Reach drafts the email, and Ring prepares the call. On support, Sage drafts the reply. On engineering, Blueprint writes the spec, Forge builds, Check reviews, and Shield looks for security risks. On finance, Ledger matches two exports and writes exceptions to a separate file. Pack drafts the commentary from numbers you provide. The finance page lists founders running their own books, so the sign-off can be you.

How one founder manages several squads

Idea #27 in the same post is "Workday for agents (onboard, manage and offboard your digital workers)".

The roster control on this site is smaller than that sentence. You import a squad onto a mission. The squads hub says a template is a starting point: keep the squad as it ships, add or remove agents, or drop the ones this mission does not need. The agent is on the mission, or it is not. That is the onboard and the offboard these pages describe. They do not describe contracts, access reviews, or an exit checklist for a digital worker.

Each mission has goals you write, and a folder you point it at. Outreach can sit in one folder. The repo can sit in another. The goals are the path: who the work is for, what done looks like, and what must not happen. The Mission Commander creates tasks on the task board and assigns each one to the agent whose role fits, only inside those goals. The commander does not get to widen that path. The product page titles the board section "One Board. Every Mission." A go-to-market card and a support card both end in Review. You approve them one at a time.

Steering is the approval step. Add a task on the board, or correct the Mission Commander in the group chat, when a draft should not go out. The go-to-market tutorial is explicit that the squad has no send cap, no suppression list, no auto-pause, and no kill switch. You write exclusions into the goals, and you withhold approval. There is no pause button to claim beyond that review step. A send waits because you have not approved the card.

What you can see before you approve

Idea #8 in that post is "Datadog for agents (see every step your agents took)".

The record here is the board and the file. The squads hub says a squad splits a goal into task cards so you can see who is doing what and review each result. In Progress means a bot is on it, editing files on your machine and posting what it found as it goes. Review means the work is waiting on you: read the diff, approve it, or send it back with a note. The product page says you see every edit on the file itself, with the bot name next to it. A card in Done keeps the bot name, and how long ago it happened.

That is not a trace of every model step. These pages do not describe a log explorer. You see the card, the write-up, and the file. You read those before the work counts.

The approval line is the same shape on every squad, with a different verb.

• Go-to-market: nothing is sent or dialled until you approve it. Ring dials only after you confirm the assistant, the caller number, the call list, and the script.
• Support and operations: nothing is sent until a person approves it.
• Marketing: nothing is published until a person approves it.
• Chief of staff: no reply is sent without your approval.
• Software engineering: nothing is merged until a person approves it. Merging and deploying stay with you.
• Finance: every output needs a sign-off before it is relied on or shared. Source data is never changed.

Anything irreversible is escalated instead of executed. The product page names refunds, cancellations, bulk deletions, and the final pricing or legal call. The human makes the final call. Nothing goes out until you approve it.

A week in a one-person startup

This is a schedule built from missions already written on the squad pages. It is not a customer story.

Monday is the chief of staff. The mission on that page is to triage the inbox into needs you today, can wait, delegate, ignore. Digest drafts replies. You send the ones you mean. Prep writes the one-page brief if a meeting is actually on the calendar. Aide keeps the list of decisions waiting on you.

Tuesday is go-to-market, if that is the job this week. Scout's list comes back with a reason for each account. You delete the ones that do not fit. Reach's emails sit in Review. If the goals include a call, Ring does not dial until you have confirmed the assistant, the number, the list, and the script. Calls need a lawful basis, local calling hours, and an assistant that says it is an AI. The agents are not lawyers.

Wednesday is the product, and it is often smaller than a rewrite. One mission on the engineering page is a script that reconciles two spreadsheets. Blueprint's spec has scope, non-goals, and acceptance criteria. Forge's change waits in Review. Check ranks findings as blocker, should-fix, or nit. You merge, or you send it back. Shield does not run exploits against systems you do not own.

Thursday is support, scaled to the tickets you actually have. Sage classifies them and drafts replies. A mission on that page is the open shipping questions. You edit what a customer will see. Billing disputes, legal threats, and safety issues are escalated to you, not answered by the agent. A help article from a solved ticket waits in the same column.

Friday is the books and the ops note, if there is something to close. Ledger writes an exceptions file and does not change the export. Variance labels every forecast assumption. Pack's commentary uses the numbers you provided. Tally compiles the weekly report from notes already in the folder: progress, risks, asks. You sign off, or you do not share it.

Seven squads on Monday morning is a way to stop reading. The product page says two or three bots is the smallest setup that works, and that you can run a squad of one. Add the next function when Review is a habit. The limit is how many cards you will actually read.

Where a person still decides

The agents draft. The founder owns the sentence.

Do not ask a squad to invent customers, metrics, or quotes. The product page says a bot that does not know flags it on the board instead of inventing an answer. A forecast assumption has to be labelled. A prospect reason has to come from a page you can open. If you would not say the line to a customer, it is not done.

The app does not drive a browser through other companies' sites. If the job lives inside someone else's web app, this is the wrong tool. The work it takes is the work in a folder: a list, a draft, a spec, a reconciliation, a reply waiting for you.

Mac now, Windows waitlist open. The Mac app is available for Apple Silicon. The Windows waitlist is open on the downloads page. There is no Windows download yet.

A one-person startup still has those functions. xSquads is the team the founder runs. The founder makes the final call.

Where these facts come from

Read on 24 October 2026.

• Greg Isenberg on X, 23 October 2026: "The MOST obvious (but REALLY good) STARTUP ideas right now". Quoted here in his wording: idea #3, "Y Combinator for 1 person companies (back solo founders running a team of agents)"; idea #8, "Datadog for agents (see every step your agents took)"; idea #27, "Workday for agents (onboard, manage and offboard your digital workers)".
• The squad pages linked above: agent counts, approval lines, missions, and audiences.
• xSquads for Work: the board, Review, irreversible work, and the note that two or three bots is the smallest setup that works.
• Squads: a template is a starting point, you can add or remove agents, and you can see who is doing what.
• How to run a GTM squad: no send cap, no suppression list, no auto-pause, and no kill switch. The stop is approval.

Frequently Asked Questions

What is a one-person startup?

A one-person startup is a company one founder runs. Finding customers, answering support, shipping the product, keeping the books, and clearing the week all sit with that person. Greg Isenberg's 23 October 2026 post names the funding version of this idea: "Y Combinator for 1 person companies (back solo founders running a team of agents)". xSquads is the team of agents the founder runs.

How do squads map onto a one person company?

Import the squad that matches the job. Go-to-market drafts prospecting and outreach. Customer support drafts replies. Software engineering specs, builds, and reviews code. Finance reconciles records and drafts the pack from numbers you provide. Marketing, operations, and a chief of staff cover publishing, the recurring admin, and the inbox. Each squad works on your files, and finished work waits in Review until you approve it.

Is xSquads a fund or an accelerator?

No. xSquads does not invest, run a batch, or take a stake. It is the desktop app where a solo founder imports squads, sets the mission goals, and approves what goes out.

How does one founder manage several squads?

Import a template for each job, keep the agents that mission needs, and remove the rest. Each mission has goals, a folder, and a task board. The Mission Commander assigns cards inside those goals and does not get to widen the path. You steer by adding a task or correcting the commander in the group chat. Cards stop in Review. The stop is your approval.

What can I see of the work before I approve it?

The shared board shows who is doing what. A card in progress posts what the bot found as it goes. Review has the change written up, and the edit sits on the file with the bot name beside it. You read that before the work counts. The site does not describe a separate trace of every model step.

What should the founder still decide?

Approve every send, post, call, and merge. Keep refunds, cancellations, bulk deletions, and the final pricing or legal call. Finance output does not change source data, and it needs your sign-off before you rely on it or share it. Do not ask a squad to invent customers, metrics, or quotes.

Related Reading

• Best AI agents for small business and solo founders: the product shortlist this page does not replace.
• How to run a GTM squad, how to run a support squad, and how to run an inbox squad.
• AI agent examples: one week of founder paperwork, with the review still yours.
• xSquads for Work and the squads list.

Raina Rufus, Founder of xSquads

xSquads for Work is free. Join the waitlist and we'll send the download link.