Every research tool we've used stops at the same place: the insight. A finding gets written up and nodded at. Then the tool's job is done, and the hard part hasn't started.
We think the insight is the halfway point. The first three posts described how AskEngine listens and remembers. This one is about what listening is for, and it's the only post here written mostly in future tense. That's deliberate. This is where we're going, and we'd rather say so plainly than pretend it's all built.
The loop, in full
Here's the shape we're building toward. You hear from users, constantly. What they say becomes a record with sources. The record surfaces a pattern and proposes what to do about it. Someone — or something — acts on it, and the product changes. Then, and this is the step everyone skips, you go back to the people who raised it and ask whether the change worked. The answer lands in the record, and the loop starts again, slightly smarter.
Most companies technically have this loop. It's made of people remembering to do things. A PM pastes a quote into a ticket. The ticket loses the link to the person who said it. The fix ships months later, nobody tells the customer, and nobody checks whether it fixed anything. Each handoff leaks a little truth, and by the end the connection between "a user said this" and "we shipped that" survives only in someone's head.
How far it runs today
AskEngine already pushes past the insight, a little. Signals carry the evidence for a pattern and the evidence against it. The system proposes next steps instead of waiting to be asked, and the inbox brings you those moments as cards: this study has heard enough, this belief just got contradicted, this recommendation is ready for a decision. You accept or you dismiss, and the record remembers which. Everything in it is also reachable over API and MCP, so the answers don't live only behind our UI.
Then the loop reaches the edge of the app and stops. You accept a recommendation, switch to your issue tracker, and the trail goes cold. That edge is what we're building against.
Where it goes next
The near-term work is making the handoff mechanical instead of human. An accepted recommendation will be able to open the issue itself, in Linear or GitHub, carrying its evidence with it — so the engineer picking it up reads the actual quotes, in the users' own words, right there in the ticket. When the fix ships, the signal will know, because status will flow back the other way. Digests will bring the week's movement to where your team already talks, instead of waiting to be visited.
Interviews won't stay the only ear, either. Support tickets say things interviews never will, and they should land in the same record, feeding the same signals, held to the same rule about sources.
The step we care about most comes after shipping. Once a change goes out, AskEngine will go back to the people whose words justified it and ask whether it worked. Their answer gets recorded against the original belief. That's the difference between "we shipped it" and "it's fixed," and today almost nobody measures it, because doing so by hand means re-scheduling the exact people you talked to months ago. A system that remembers who said what can just ask.
And then there's the reason we keep mentioning MCP. More software is getting built by agents, and agents have a problem humans don't: no hallway conversations, no accumulated feel for the customer. Whatever they know about your users, they have to be told. A queryable belief layer is how you tell them. Before an agent writes the fix, it checks what users actually said. After it ships, it checks whether they're happier. The loop stops needing anyone to remember anything.
Honest accounting
Where each piece stands, in one breath: the record, the signals, the inbox and the MCP access are shipped and running in the beta. The issue-tracker handoff and the digests are next. The validation step — going back to ask — is the destination, and it's the part we'd build first if building worked that way.
These field notes exist because we're doing this in public. If you'd rather watch the loop close from the inside, the waitlist is open, and it starts — of course — with an interview.