User research expires. The report you wrote in March describes the users you had in March. By September the product has changed, the users have changed, and that report reads like a postcard from a place that no longer exists.
Most teams know this, and deal with it by doing research again. A new round of interviews, a new deck, a new doc that starts expiring the day it's finished. Research stays a project: something with a start date, an end date, and a long quiet gap until the next one.
Continuous user intelligence is the alternative. Instead of studying your users in bursts, you keep a living record of what they believe, and you keep it current. Every new conversation feeds it. Every claim in it traces back to something a real person actually said.
That's the whole idea. The rest of this post is what those words mean in practice.
Continuous means cheap enough to never stop
Research happens in bursts because interviews are expensive. Someone has to recruit, schedule, host, take notes, and write it all up. When each conversation costs hours, you ration them.
Shrink that cost enough and the rhythm changes. You talk to users after every churn. After every launch. After a support ticket that made no sense. Listening stops being a quarterly event and becomes a background process, like analytics. Analytics never sleeps, and nobody finds that strange. Conversations should work the same way.
The volume matters for a second reason. Three interviews give you anecdotes. Thirty give you patterns, plus something rarer: the exceptions that complicate the pattern. You only see those when you can afford to keep looking after you think you have the answer.
Intelligence means sourced and honest
A folder of transcripts is continuous too. It's also useless. What separates intelligence from a pile of notes comes down to a few properties.
Every claim has a source. "Users find onboarding confusing" is an opinion until you can click through to the four people who said so, in their own words. Summaries drift. Quotes don't. If a belief can't show its receipts, it shouldn't be in the record.
The record updates itself. Beliefs about users go stale silently. A good system notices when new conversations contradict an old conclusion and says so, out loud, instead of letting the old conclusion sit there looking confident.
It admits gaps. Sometimes the honest output is "we haven't heard from enough churned customers to know." A system that always has an answer is guessing. One that can say "go collect more evidence here" is doing its job.
You can ask it questions. A record you can only read top to bottom is an archive. The point is to ask "what do enterprise users say about pricing?" and get an answer with sources, whether you ask through a search box or your tools ask through an API.
Old reports decay. This compounds.
Here's the practical difference. A research report loses value every month, because it's a snapshot. A sourced record gains value with every conversation, because patterns need history. You can't notice that a belief is shifting unless you remember what it used to be, who held it, and when. The twentieth interview is worth more than the first, since it lands on top of nineteen others the system still remembers.
This is why we think the project model is ending. A snapshot, however good, can't compound.
Where AskEngine fits
AskEngine is our attempt to build this end to end. You share a link to invite someone to an interview, and each conversation is cheap enough to run week after week instead of saving them up for a quarterly burst. Everything said flows into one record: patterns surface as signals, each backed by quotes you can click, with contradictions flagged and gaps named. The same record is reachable over API and MCP, so your agents can check what users believe before they act.
You could assemble something similar yourself from transcripts, spreadsheets and effort. Some teams do, for a while. The hard part was never any single piece. It's keeping the loop running every week while you also build a company. That's the part we wanted to make automatic.
If your team still learns about users in bursts, start with one always-open interview link. Let it run for a month. The difference between a snapshot and a record becomes obvious the first time you get to ask, "wait, when did they start saying that?"