FOMAT: When Your AI Agent Follows You Home

There is a particular kind of productivity theater that the tech industry has perfected over the decades: take a structural problem about how work bleeds into life, rename it as a personal workflow challenge, and sell the solution as empowerment. I want to be careful about whether Cmd+Ctrl falls into that category — because the tool itself is thoughtful, and Michael Richman's diagnosis of the underlying problem is accurate. But the framing deserves more scrutiny than it's getting in the developer tooling conversation.

Richman, who leads engineering teams at Bitly and co-leads the company's AI coding tools strategy (per his own presentation), coined the term FOMAT — Fear of Missing Agent Time — at the AI Engineer conference this week. The concept is genuinely useful: you kick off a coding agent, walk away, and discover forty minutes later that it stopped after two minutes to ask you a clarifying question. It's been blocked ever since. As Richman puts it, "the longer the agent waits for you, the more agent time that you have missed." His solution is Cmd+Ctrl, a system that sends push notifications to your phone or watch when an agent needs input, lets you respond or launch new sessions remotely, and aggregates all your active agent sessions — across Claude Code, Cursor, Codex, Gemini CLI, and others — into a single dashboard.

The architecture is worth understanding because it explains what's genuinely differentiated here. Rather than hooking into any single agent's API, Cmd+Ctrl runs a daemon alongside each agent platform. Those daemons report to a shared control plane, which the mobile and web UIs then query. The result is platform-agnostic by design: Claude Code running on your Mac and a Codex CLI instance running in a cloud VM both surface in the same interface. The daemon layer is open source, meaning developers can wire in whatever agent framework they're building with. Richman demonstrated this during his talk with live session mirroring between a terminal and an iPhone simulator — start a session on either end, continue it on the other, seamlessly.

The standup dashboard feature deserves particular attention as agent management tooling matures. Rather than requiring a developer to read through every session thread to reconstruct what happened, Cmd+Ctrl generates a summary of recent sessions from the last several messages of each. Anyone who has returned from a meeting to find three agents in various states of completion — one blocked, one finished, one apparently hallucinating a file structure — will understand the appeal immediately.

Richman is also honest about the current limits of agent autonomy in a way that the broader industry marketing is not: "We all want to believe that our agents are low touch and high autonomy. But we all know the truth, right? It's back and forth, it's babysitting and you cannot predict when you're going to be needed for input." He notes that as agent task durations extend — from minutes to hours to eventually days — the unpredictability of when human input is required becomes the central workflow problem. The overnight agent paradigm already pushes this into territory where checking in periodically simply doesn't work.

All of that is real, and Cmd+Ctrl addresses it competently. Here is where I want to slow down.

Richman closes his talk by observing that "this paradigm of the always available agent is actually highlighting the value and importance of time away from your agents." He cites the cognitive load of managing multiple agent sessions and argues that we need systems enabling developers to reach their agents during breaks — "wherever we might be and whenever that happens." The framing presents Cmd+Ctrl as the thing that makes genuine rest possible, by removing the anxiety of missing critical agent moments.

That framing is worth examining from a direction Richman doesn't take. A tool that ensures you can respond to your agents from anywhere is not the same thing as a tool that ensures you only have to respond when you choose to. Those are different propositions with different implications, and the distinction matters enormously depending on your employment context.

The right-to-disconnect question has actual legal teeth in some jurisdictions, even if U.S. policy has largely declined to engage with it seriously. France's droit à la déconnexion, codified in the El Khomri labor law reforms of 2016, requires companies with more than 50 employees to negotiate policies on after-hours communication. Italy, Spain, Portugal, and Belgium have followed with their own frameworks. The EU's Work-Life Balance Directive creates broader floors around worker contact expectations. California has seen repeated legislative attempts — none yet successful — to establish similar protections. The EU AI Act, for its part, concerns itself with AI system risk classification and transparency obligations, but creates no framework for questions about human supervisory burden or how that burden is distributed across a worker's waking hours.

None of these frameworks were designed with the agent-supervision relationship in mind, and that gap is worth naming explicitly. When a developer is nominally off-hours but their agent — running on company infrastructure, against a company codebase, on company cloud resources — sends a push notification asking for a decision, what is the legal character of that interaction? Is it work? Under most employment law, if you are performing a function that advances your employer's interests, the answer is yes — regardless of whether you're in your pajamas or on a Sunday hike. The notification didn't come from a colleague who overstepped; it came from an automated system that your employer sanctioned and that your job performance may increasingly be measured against.

Richman's presentation is aimed at engineering leaders and individual developers with substantial autonomy over their own hours. For that audience, the "negotiate your own availability" framing is plausible. But software development at scale doesn't only happen among senior engineers at tech-forward companies who set their own schedules. It happens among junior developers under delivery pressure, contractors whose billing arrangements don't cleanly separate hours, and workers in geographies with weak labor protections who may experience "you can respond from anywhere" as "you are expected to respond from anywhere." The tool doesn't create that pressure — industry culture did that long before AI agents arrived — but tools that make constant availability technically frictionless tend to normalize it as well.

There is also a narrower accountability question that nobody in this space has answered adequately: when an agent running unsupervised takes a consequential action — commits code, makes an API call, modifies a production configuration — and the developer was unreachable or chose not to respond to a notification, who bears responsibility? Current software development contracts and employment agreements were not written with this scenario in mind. Neither were most company AI use policies, which tend to address outputs rather than autonomous process decisions made by agents between human check-ins. The trajectory toward fully autonomous coding systems makes this less hypothetical by the month.

Richman acknowledges, candidly, that Anthropic had recently released its own mechanisms for remote agent interaction and that Cursor shipped something in this space close to the time of his presentation. He's building in a field that the major platform vendors are also entering, which creates obvious questions about how long an independent control plane layer remains differentiated. But that competitive dynamic is secondary to the more durable question his work surfaces: as the human-agent supervision relationship becomes a standard part of software work, what structures — legal, contractual, technical — govern how that supervision is distributed, compensated, and bounded?

Richman's answer is a well-built tool that gives you the phone notification and the watch tap and the standup dashboard. That is a meaningful contribution to a real problem. The question it leaves open isn't whether the tool works. It's whether "you can now respond from anywhere" will be experienced, by most of the people subject to it, as freedom or as a different kind of constraint — and whether anyone in a position to set policy is paying attention yet.

---

Samira Barnes is Buzzrag's tech policy and regulation correspondent.