LaterOn Blog

Try LaterOn!
LaterOn v2: From Newsletter Reader to Agentic Email Platform
Product
Agentic Email

Sep 16, 2026   •   By M Gilang Januar

LaterOn v2: From Newsletter Reader to Agentic Email Platform

LaterOn began with a narrow promise. You got a separate address for your subscriptions, we summarized what arrived, and you read it on your own schedule instead of letting it interrupt your day. That still works and it is still part of the product. But it solved the smaller half of the problem: reading was never the expensive part of email. Answering was.

v2 changes what LaterOn is for. You describe an email job once, in plain language, and an agent runs it inside your mailbox. It watches for the emails that match, replies with your context and in your voice, chases the ones worth chasing, and closes the conversation the moment your goal is met.

The new goal: fewer threads you have to touch

Most inbox tools are built to help you get through your email faster. Sorting, snoozing, summarizing, bundling: all of it still ends with you writing the reply. The work only looks smaller.

So the goal for v2 is not a tidier inbox. It is a smaller number of conversations that need you at all. Everything below exists to serve that one measure.

What a workflow actually is

A workflow is four decisions you make once:

  • A trigger. Conditions on subject, sender, recipients, cc, body, labels, or attachments, matched with all or any of them. On top of that you can add a plain-language gate such as "only when they are actually asking about pricing", which is judged after the conditions pass, so the model never has to read mail your rules already ruled out.
  • Instructions. How to answer. What to say, what to send, what to never promise. This is your writing and your rules, not a template.
  • An objective. What finished means. "They have our pricing and told us how many seats they need." On every turn the agent reports whether the objective is met, still pending, or unreachable, and the run ends when it is met.
  • Limits. How many times it may reply in one conversation, how many conversations it may touch per day, and whether it sends or only drafts.

One action per turn, with a reason

Each time a workflow looks at a thread it chooses exactly one action: reply, skip, complete, or escalate. Every choice carries a one-sentence reason written for you to read, and it lands in the run timeline next to the message it sent.

Escalate is the interesting one. When the thread needs a decision the agent does not have, it stops and hands the conversation back to you rather than inventing an answer. An agent that knows when to stop is worth more than one that always has something to say.

Follow-ups that stop themselves

A reminder is armed when a reply goes out and cleared the moment the other person answers. Nobody gets chased for a message they already replied to. You choose how long the silence has to last, how many times it is worth trying, and when to give up on the thread entirely.

Guardrails, because it sends real email

  • Every workflow is created paused. You turn it on.
  • Draft mode writes the reply into the thread and waits for you, so you can watch a new workflow for a week before letting it send.
  • Caps on replies per conversation and conversations per day are required, not optional.
  • Newsletters, no-reply addresses, and senders on your blocklist are never answered.
  • Everything the agent sends is in your mailbox, with the reason it sent it. You can read every word it wrote in your name.

The newsletter reader is still here

Nothing was taken away. The mailbox, the digests, the summaries, and the reading experience all stayed. If that is all you want from LaterOn, keep using it exactly as you do now: workflows sit next to it and only ever touch the threads you point them at, and they never answer a newsletter.

Where this goes

The bet behind v2 is that the useful unit of email automation is not a rule and not a chat window, but a job with an owner, an objective, and a limit. Rules cannot tell whether your goal was met. A chat window cannot act while you are asleep. A workflow can do both, and then tell you what it did.

Describe one job you would rather not do again, and let it run.

Back to Blog