Hand Marvin a question. Go do something else.

You ask ChatGPT a question. The cursor blinks. The answer streams in word by word. You watch it write. After thirty or sixty seconds, you have a response. You read it, ask a follow-up, watch the next answer stream. You're in the conversation.

This works great for some kinds of work. It works terribly for others.

The other kind is research that takes longer than you can reasonably wait while staring at a screen. Reading a 60-page S-1. Pulling together a comparative analysis across five companies. Surveying a sector you don't follow closely. These are 5-to-15-minute jobs even for an AI. You can't sit there watching them stream, and you can't really do other work productively while a chatbot tab is in your peripheral vision waiting for you.

The asynchronous workflow is the better fit. You hand off the question, and then you actually go do other things, and the result arrives in your inbox when it's done.

What that looks like

Tuesday morning. Coffee, calendar review, three emails you can clear in five minutes. Then you remember: you need a deep brief on Anthropic before your partner meeting at 11. You email Marvin:

To: marvin@ccmarvin.com

Hi Marvin,

Do a deep research report on Anthropic: founders, traction since the last funding round, biggest risks, and where they sit relative to OpenAI and Google in enterprise share. Cite primary sources where you can.

A reply lands in under a minute, then you put your phone down:

From: Marvin marvin@ccmarvin.com

Hi Tom,

On it. This is a deep one, so it'll take a few minutes. I'll send the full report to this thread the moment it's ready. No need to wait around.

Sincerely,
Marvin

You go back to clearing emails. You take a call. You write your standup. At 8:47 the report lands in the same thread: a structured, multi-thousand-word brief, citations to primary sources, sections matching what you asked for. You skim it on your phone walking to your next meeting. By the time you sit down at 11, you've read it twice and added your own marginalia.

You did seven other things in the time Marvin spent on this. That's the workflow.

Why this matters more than it sounds

The real value of asynchronous research isn't that the AI is faster. It's that your time isn't blocked.

A 15-minute synchronous task, even one where you're "just watching it stream," eats real focus. You can't context-switch productively while waiting, so the 15 minutes of AI work becomes 15 minutes of your attention plus context-switching cost on either side: call it 25-30 minutes of drag for a 15-minute job.

Handed off async, the same task costs 30 seconds to write the prompt, zero attention while it runs, and three minutes to read the result. Stack that across five or six research tasks a week (briefs before partner meetings, sector reads before strategy calls, diligence before investment committee) and the difference is real hours back in your day.

Why your report survives our deploys

A research run takes a few minutes. We ship code several times a day. A naive server would restart mid-job and lose your work right when a deploy lands.

We refused to let that happen. No Marvin user should ever have to ask "did your deploy cancel my research?" The answer needs to always be no.

So research jobs persist through restarts. When we restart, we pick up any in-flight research where it left off. From your side, the deploy is invisible: the report still arrives when it would have.

What about follow-ups?

Once you have a report, you usually want to dig into one section or ask something that builds on it. Reply to the same thread and Marvin builds on the prior report rather than starting over. The follow-up is faster and more coherent because the context is already established. (More on that here.)

How to try it

The next time you have a research question that would normally take you an hour with browser tabs, email it to marvin@ccmarvin.com instead. Then close the tab and go do something else. The report will be waiting when you check back.

The first time you do this — and actually get the time back for other work — is when you realize how much of your day used to disappear into research you were doing yourself.