Post 4 of 5 — The Handoff series

This is the one we're most excited about, so let's just show you the flow.

You're in Earmark. You type: "Write me an email recapping the Insta meeting to send to Mark." Earmark uses its memory to find the meeting, pull what matters, and draft the email - right there in chat.

The email's good. Maybe you tweak a line. And then, instead of copying it into another tab and re-explaining yourself to yet another tool, you click one button:

Continue in Claude. Or Continue in Codex.

Your local agent opens, already seeded. There's a short preamble - you're receiving an Earmark handoff - then the work itself: the email, the meeting it came from, the context to run it. You hit go. Switch to your inbox, refresh, and the draft is sitting there, ready to send.

You didn't connect Earmark to your email. You didn't set up an integration. You used the agent you already had.

What's actually happening

Behind the scenes, the handoff is simpler than it sounds - and that's a feature, not an apology. It's a deep link. But instead of deep-linking into a tool like Linear or a doc editor, we deep-link into your agent, carrying instructions to execute the work.

That distinction is the whole idea. Once you're handing off to an agent, you don't need to target individual tools anymore. Your agent already knows your tools. You don't tell it which integration to use - you tell it what you want, and it figures out the rest. Ask it to file the tickets in Linear, and it checks for duplicates, confirms the project name, and does the thing. None of which we had to build.

The prompt is pre-rendered into the link, so clicking it doesn't kick off a round-trip - the instructions are already there, ready to run. And it's still editable before you send. Even when it's not perfect, the blank-page problem is gone. You're never staring at an empty box wondering how to phrase the ask. The ask is already written.

We keep the button deliberately dumb: it's always there, like a copy button, on every artifact. No guessing whether you wanted it. It just waits for you.

It goes both ways

Connect Earmark to your agent over MCP and the handoff stops being one-directional.

Your agent can pull more from Earmark. Say the handoff gives you a draft and you want it fleshed out - "add more detail, pull a key quote." Your agent reaches back into Earmark's memory through the MCP and gets what it needs. The same riffing you'd do in our chat, now happening inside Cowork.

Your agent can push back into Earmark. Reading an email in Cowork and realize you need to raise it at your next standup? Just tell Claude: "remind me to bring this up at my next engineering standup." It writes that reminder into Earmark. A week later you open the meeting and there it is, waiting on your agenda - placed there by your agent, not by you.

That two-way street is the part that's genuinely new. Earmark and the agent you already run, talking to each other, in both directions.

Why we're not building our own agent (yet)

Notice what we didn't do here. We didn't build an agent. There's no harness we had to ship to make this work. It's a deep link and a well-formed prompt.

That's not laziness - it's strategy. The labs are improving their agents weekly. Rather than race them, we piggyback on how fast they're moving. The better Claude Cowork and Codex get, the better this handoff gets, for free. We take a path almost nobody else is taking: not implementing our own agent, and being refreshingly honest about it.

It also makes the security story clean. We don't connect to your tools. We don't need broad access to your stack. We hand context to the agent you already trust, and it does the rest inside the boundaries your company already approved.

For the people who don't have a local agent set up, we'll offer a managed option that routes the same workflow to an agent we run. Same flow - you just choose where the work happens. But for most people, most of the time, the answer is going to be: locally, on the agent I already love.

So when someone says "we already use Claude Cowork" - that's not an objection. That's the reason to use us.

Last post in the series: the loops that mean you don't even have to ask.

Want early access? waitlist@tryearmark.com.