Ondago · Your agents are blocked, waiting for you

Six cases we keep seeing

Nothing here is broken. Everything here is idle.

The session is welded to one machine and one person, so it only moves while you are sitting in front of it. These are the six ways that shows up in a normal week.

The permission prompt

It asks to run a migration at 10:40. You are in a meeting until 13:00. One keystroke of work, two hours and twenty minutes of nothing.

Blocked on one yes

The decision it cannot make alone

Not an approval, a judgement. Which approach, which file is the source of truth. It waits for the only person who has read the session.

Blocked on you specifically

The silent finish

It finished forty minutes ago and nothing told you. The next task cannot start until you happen to look.

Blocked on nobody. Just idle

The stop short

You come back the next morning and it halted somewhere in the middle. You reconstruct where it got to and do part of the work again.

Work repeated, not resumed

The one-person lock

A teammate is awake, has the context, and could unblock it in a minute. There is no way to hand it over. You can pass a branch, not the work in flight.

Blocked on the fact that it is yours

The private session

The reasoning stayed on one laptop and the team got a diff. Nobody knows what the change was meant to do.

The cost lands on everyone downstream

The number nobody measures: hours your agents spent waiting on you this week.

What the research says

9%

of developer time in a controlled study went to waiting for AI, with fewer than 44% of its generations accepted.

Becker et al., METR, 2025 · arxiv.org/abs/2507.09089

Effort down,
clock time unchanged

The first controlled study of developers using coding agents found agents reduced the effort a task required, but not the time it took to finish.

Code with Me or for Me?, CHI 2026 · arxiv.org/abs/2507.08149

Worth knowing before you look it up yourself: the same METR study found developers were 19% slower overall with AI while believing they were 20% faster. Nobody has yet measured the number this page is about, which is how long an agent sat idle waiting for a person. We intend to.

How it works

One command. The session moves. You leave.

$ ondago lift

Lift the running session

Mid-run, from your terminal. No clean git state required, nothing to configure in advance, no deciding to work remote before you started.

→ relocated, not mirrored

It keeps working in the cloud

The same conversation continues on a sandbox. Close the lid, get on the train, let the battery die. The run does not notice.

● reaches you, anywhere

It comes to you when it needs you

A prompt or a finish reaches your phone, any laptop, or a browser. You answer in a minute instead of finding it parked hours later.

The work travels wholeConversation, working directory and uncommitted changes move together. Nothing to commit or stash first.

On-demand, not for goodYou do not move your life to a remote box. You move one session, when leaving matters, and keep your local workflow.

When you come backFinished, or waiting exactly where it genuinely needs you. Never stopped short in silence.

Where this goes

Once the session leaves the laptop, it stops being single-player.

No answer from you in fifteen minutes and it goes to the team channel, or to someone in another time zone who continues the session rather than restarting it.

The session becomes the record: what the agent was asked, which files it read, what it tried and discarded, who approved the migration. Review the work, not only the diff it left behind.

Your agents are blocked,
waiting for you.