Your AI Coding Assistant Doesn't Need You to Watch

Have you noticed something odd about how we use AI for coding today?

You open a terminal, type an instruction, then stare at the screen watching it output line by line. When it finishes, you glance at the result, then type the next one. If you step away to grab a coffee, you come back to find it finished ages ago, just sitting there waiting for you. Or worse — it went off track, and because you weren't watching, it changed a bunch of files it shouldn't have.

This is a synchronous interaction model. You and the AI take turns talking, one exchange at a time, neither can leave.

But writing code has never worked this way.

Your Work Isn't Synchronous

Think about how you actually work. You probably have three PR reviews open at once, waiting for CI to finish, responding to Slack messages in between, occasionally switching to another branch for a hotfix. Your attention jumps between multiple tasks rather than locking onto one.

But when you use an AI coding assistant, you're forced to become a synchronous person. You have to sit at your computer, watch it work, wait for its questions, and respond immediately. If you're in a meeting, the AI just waits. If you're commuting, it's completely unusable.

This isn't the AI's problem. It's an interaction model problem.

What If You Could Tell It What's Next, in Advance

Imagine a different scenario.

You arrive at the office in the morning, pull out your phone, and spend five minutes listing what you want AI to do today: first, refactor the form validation on the login page; then standardize the API error handling into the new format; finally, run the test suite to see if anything broke.

Three tasks, queued up, sent off. Then you go to standup.

Midway through standup your phone buzzes. A notification tells you the first task is done. You glance at it — looks good, the second task has already started automatically.

After standup you check your phone. The second task is done too, but the third one hit two failing test cases and the AI paused, waiting for your input. You look at the failures, realize the expected behavior changed, and tell it to update the test assertions. Phone goes back in your pocket.

Before lunch, all three tasks are complete. You didn't even sit down at your computer all morning.

This Isn't Automation — It's Delegation

There's an important distinction here.

Automation means you define rules and the machine follows them exactly, no judgment required. CI/CD is automation. Linting is automation.

But AI writing code isn't automation — it makes judgment calls, and its judgment isn't always right. So you can't fully let go, but you also don't need to watch every second.

The more appropriate model is delegation: you tell it what to do, it goes and does it, it comes back to ask when something is uncertain, and it tells you the result when it's done. Just like assigning a task to a reliable colleague.

For this model to work, you need a few things:

You need to plan ahead. Not one instruction at a time, but multiple tasks queued up at once. So when the AI finishes one, it automatically starts the next without waiting for you to be online.

You need to course-correct mid-flight. If you notice the AI is heading in a slightly wrong direction, you should be able to say "change course" without canceling and starting over. Canceling means losing context, and all you really wanted was a nudge.

You need to see its plan before it acts. Especially for big changes — refactoring a module, upgrading a dependency — you don't want the AI to just dive in. You want it to say "I'm planning to change these three files, here's the logic," and only proceed after you confirm.

You need to know what's happening without being at the screen. Task finished, error occurred, decision needed — this information should come find you, not wait for you to remember to check.

Your Phone Is a Natural Async Control Panel

Why the phone?

Because the phone is already your device for handling asynchronous information. Email, messages, calendar reminders — you don't stare at them constantly, but they alert you when needed. An AI coding assistant should work the same way.

Your computer is where the coding happens. The AI runs there, reads your code, modifies your files, executes your tests. But control and supervision don't have to happen on the same machine.

Reviewing the AI's execution plan on your phone and confirming it. Getting a notification on the subway that last night's queued tasks are all done. Quickly adjusting the direction of a running task between meetings. In all these scenarios, the phone is more natural than the computer.

Synchronous Interaction Isn't Going Away

I'm not saying sync mode is useless. When you're deep-debugging a bug, going back and forth with the AI on an approach, or doing exploratory work that needs instant feedback, synchronous mode is still the best choice.

But not all programming work requires deep engagement. Much of the time you know exactly what needs to be done — it's just tedious to execute. These tasks don't need you watching the AI do them. You just need to know it's done, and done correctly.

Async mode fills in that piece of the puzzle.

This Is What 1.6.5 Does

The new features in Tactic Remote 1.6.5 — plan mode, prompt queue, steer, push notifications — aren't separate features. They're different facets of the same idea: letting you control your AI coding assistant asynchronously.

Queue it up, review the plan, confirm execution, get notified. Then go do something more important.

Your AI coding assistant doesn't need you to watch. It just needs to be able to reach you when it matters.