Company profile · 4 min read
Agent FM: reviewable AI work across files, browsers and agent sessions
Agent FM combines agent supervision and cross-app work automation with visible progress, approval gates and reviewable results.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
Agent FM is trying to turn agent supervision from a row of terminal sessions into a work queue with a visible handoff and approval moment. Its public positioning now reaches beyond coding agents: founders, operators and teams can describe work across files, browsers, inboxes and business apps, then review the result before anything goes out.
What it does
The current homepage says Agent FM plans work across connected tools, brings the finished result back to the same chat and asks for sign-off before sending or submitting. Its examples include invoice reminders, inbox triage, CRM updates, deck refreshes and spreadsheet reconciliation. The page names Gmail, Outlook, Calendar, Slack, QuickBooks, Salesforce, HubSpot, Docs, Sheets and other services, and says it runs on macOS 14+ and Windows.
The original YC description is narrower: a desktop group chat to hear and steer Claude Code, Codex and other coding agents, with live text and audio updates about progress, blockers and decisions. The product story has expanded, but the common promise is the same: see what an agent is doing, interrupt when needed and keep outward actions behind review.
Why I’d look closer
The founder context matches the infrastructure problem. YC describes Aniket Deshpande as previously working on applied AI at Meta Superintelligence Labs and Mugdhaa Patankar as previously leading AI for developer productivity at Snowflake, with earlier work at Twitch and AWS. The launch post also reports 410 engineers using Agent FM and 60-plus paying users; those are company-reported figures, not an independent adoption measurement.
The advantage is the approval-centered workflow. A founder can delegate a concrete outcome without giving up the final send, submit or purchase step. The tradeoff is breadth: an agent that can touch inboxes, files and signed-in sites needs precise scope, clear logs and connector-specific failure handling. “Works wherever your work happens” is a useful promise to test, not a reason to grant blanket access.
What I’d ask
Which connectors are native, which are browser-driven and which require extra setup? Can a user preview every external change and revoke a session mid-run? How are credentials, local files, action logs and model-provider data separated? What happens when a site changes or the agent reaches an irreversible step?
My editorial take
Shortlist Agent FM if repetitive cross-app work is eating the founder or operations team’s attention and a reviewable handoff is more useful than another chat window. Start with read-heavy, reversible workflows, keep approval on all outward actions, and measure time saved against correction time. The product’s differentiation is control around execution, not merely a longer list of integrations.
Quick facts
| Field | Sourced detail |
|---|---|
| Product | Desktop agent-supervision and cross-app work-automation surface |
| Buyer | Founders, chiefs of staff, operations, sales and engineering teams |
| Availability | Homepage says start free with no credit card; detailed plan pricing is not published |
| Control claim | Site says outward actions wait for user approval and steps remain reviewable |
| Main question | Can the team delegate useful work without losing scope and approval control? |
Sources checked
| Source | Checked |
|---|---|
| YC company profile | 2026-09-19 |
| Agent FM homepage | 2026-09-19 |
| Agent FM blog | 2026-09-19 |
Cohort context
Agent FM is listed in Summer 2026. In our 2026-09-18 directory snapshot, 119 of 232 listed companies in that cohort have YC’s primary industry label B2B (51.3%). This is a current-directory comparison, not an original intake count or a performance ranking. Nine-cohort dataset.
Public website snapshot
Observed 2026-09-19T16:18:31.471Z in raw homepage HTML. This records visible metadata and advertised links, not agent execution or product quality.
| Signal | Homepage observation |
|---|---|
| Product description metadata | Observed |
| Canonical link | Observed |
| H1 or H2 heading | Observed |
| Typed structured data | Observed |
| Docs/developer link | Not observed in this response |
| Pricing link | Not observed in this response |
| llms.txt link | Not observed in this response |
| Markdown alternate | Not observed in this response |
Public observations · Collection method. Missing links here do not establish that a capability or file is absent elsewhere.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗