00 How it works
The loop, in full.
The homepage version of this is four lines. This is the long one: what it reads, how it decides what matters, what it does about it on its own, and the exact point where it stops and waits for you.
01 It reads
The context is the product.
Everything downstream depends on this part, and it is the part a chat window cannot do, because a chat window only knows what you thought to paste into it.
Your records
Customers, jobs, quotes, tickets, whatever your business actually tracks. Not a copy of them in a separate index somewhere, the records themselves.
Your pipelines and the stages in them
Where each thing has got to, how long it has been sitting there, and what it is linked to that has not happened yet.
What happened, and when
Every event against a record. A stage change, a file that landed, a form somebody filled in, a reply that came back, a run that finished.
What it has learned about how you work
Facts it holds about the business, each one carrying where it came from: something you stated, something you confirmed, or something it inferred and has not had confirmed yet.
02 It decides
Knowing what matters is the hard part.
Acting is easy. Software has been able to act for decades. What nobody automated is the judgement in front of it: out of everything that could be done today, which of it is worth a person, and which of it is worth you specifically.
A general manager agent holds the whole loop rather than any one job in it. It is the thing that notices the proposal has been quiet, decides that matters more than the tidy administrative task also sitting there, and puts the right specialist on it.
You are not managing a roster of bots.
That distinction is the reason there is one thing to talk to rather than twenty. You say what you want in the language you would use with a person, and the dispatching happens underneath. If a job needs three specialists, that is not your problem to solve.
03 It does
Then it goes and does the work.
Each agent has its own environment in the cloud, with a browser, a shell and a file system of its own. That is what separates producing from suggesting. It can go out and read a source, work figures against your records, build a real document, and file the result against the record it belongs to.
Set a job to repeat and it runs while the office is closed. Give it a goal instead of a task and it keeps working until an independent check agrees the goal is met, or until the attempt cap you set runs out, which it reports as a timeout rather than dressing up as a success.
When the provider underneath all of this went down for most of a day, the scheduled work held its slots and picked itself back up. Nothing was lost and nothing ran twice.
- Opened a browser and read the company site
- Pulled the last three jobs off the record
- Wrote the brief as a document
- Filed it against the account
- Staged the email. Did not send it.
04 It stops
And then it waits for you.
Everything above happens so that this part can be short. What reaches you is a queue of decisions, sorted by what ignoring them costs, with the work already done underneath each one.
Things to approve
Something it wants to send. You read the whole thing as it will go out, then approve it, edit it first, or skip it.
Things only you can settle
A question it genuinely cannot answer from the records, put to you with the options it thinks are reasonable and what each one would mean.
Things it noticed
A recommendation with no deadline on it. Worth a look when you have a minute, and it will still be there when you do.
05 It learns
Then the loop closes and gets shorter.
What you did with each item is information. Approving something as it stood, editing it first, or skipping it entirely all tell it something different, and it keeps them apart.
Corrections stick. If you fix a fact it holds, that correction is what it uses from then on, and the fact now carries your name rather than its own guess.
Answers become rules. A question you settle once can be turned into the standing rule underneath it, and every agent has that from then on rather than asking you again.
And the work you keep approving the same way is the work that earns its way up the ladder, which is how the queue in front of you is meant to get smaller rather than larger.
Executive expertise that executes.
Point it at the loop that keeps slipping.
ExecFlow is in beta with a small group of early testers. The useful version of this conversation starts with the thing in your business that keeps falling through, not with a feature list.