Ask Marvin a follow-up: research that remembers across days

The thing we kept hearing from users running deep research with Marvin was a version of this:

"I got a great report on Monday. On Wednesday I want to ask a follow-up that builds on it. Right now I have to paste the relevant section back into a new email so you have context. Can you just remember?"

The answer used to be no. Each deep research run started from a fresh context window, so a follow-up either got answered shallowly (no prior research to lean on) or re-ran the whole thing from scratch (slow and expensive).

We just shipped a better path. Marvin's deep research now remembers your prior reports on the same thread, and builds on them when you ask follow-ups.

What it looks like

Say you got a full report on a company on Monday. On Wednesday you reply to that same thread:

To: marvin@ccmarvin.com

Hi Marvin,

Can you dig into the regulatory section of Monday's report? Specifically what's happening with the EU AI Act and how it would affect this company's go-to-market in Europe.

The report that comes back doesn't restart from zero:

From: Marvin marvin@ccmarvin.com

Hi Maya,

Building on Monday's brief: the EU AI Act's high-risk classification is the live issue for them. Their core product would likely fall under the "limited risk" tier, so the conformity-assessment burden is lighter than the headlines suggest, but the transparency obligations still bite.

Three things that change the EU go-to-market specifically: documentation requirements add roughly a quarter to enterprise sales cycles, the August 2026 enforcement date sets the real deadline, and their current data-residency setup (noted in Monday's brief) already covers most of what auditors will ask for.

I didn't re-run the company overview since that's unchanged from Monday. Want me to map this against a specific competitor's EU posture next?

Sincerely,
Marvin

The new report references Monday's work by name, introduces only the new material the follow-up needed, and runs roughly half the length because it isn't re-establishing baseline context. It's faster too: typically one to two minutes instead of three to five, because the "understand the company" work already happened.

What it's built on

Marvin chains a new research request to a prior completed one. When he recognizes a follow-up on the same thread, he passes the earlier research forward as context, so the new run starts where the last one left off.

If there's no prior research on the thread, or the prior one is too old to be useful, he just runs normally. If the follow-up departs from the original topic, he mentions the shift and proceeds with a fresh approach.

There's also a fast path for follow-ups that need no new research at all. Ask a clarification ("what was that company's last funding round again?") and the answer is already in the report Marvin wrote. You'll get it in seconds rather than minutes. The response time tells you which path ran.

What changes for you

Follow-ups on a research thread now feel like a conversation. Ask for more depth on a section. Ask for an angle the original missed. Ask Marvin to compare the company you just researched against another. Each reply builds on what came before.

If you want a fresh run instead of a follow-up, start a new email thread. Threading is now a real signal: same thread means "build on what we have," new thread means "start over."

How to try

Send a deep research question to marvin@ccmarvin.com. When the report lands, reply on the same thread asking for more depth on one section. Watch the next report pick up where the first left off instead of starting over.