I love intro chats with interesting people.
A few weeks ago I had an awesome one with another founder from a warm introduction. Ten minutes in, we were both deep in the weeds on building right now: what feels possible, what feels insane, and the strange fun of trying to turn massive uncertainty into a company.
At the end, we agreed to reconnect. He was going to review his network and send over a few more warm intros. I sent a thank you to the person who connected us, made a note, and went heads down.
Newsletter. Customer calls. Product. Marketing. More intros. The usual founder mix.
Two weeks disappeared.
Then one morning I started a new agent session, and it told me one of my active goals had flipped to at risk.
"15 days quiet from the counterparty's side."
I had given the intros two weeks to land, and the timer had run out.
I forgot.
The product remembered.
Last week I wrote about missing the team.
More specifically, missing the room. The daily grind at Apple where someone would poke at your thinking, catch your blind spot, or ask the annoying question you secretly needed.
What I left out is that I've been building toward that missing function.
The room cannot be replaced, and it really shouldn’t be. There is magic in real people paying attention together.
But teams also do something more basic and practical.
They catch each other's misses.
A dropped commitment. A loose follow-up. A little thing everyone thought someone else owned. A customer thread that went quiet. A promise that was made in a meeting and then drifted into fog.
Good teams notice the silence.
Of course not every silence means something, but enough of them do. And when you are solo, busy, or buried in the week, you do not always have another person around to say, "Hey, whatever happened with that?"
I keep coming back to that gap.
Question-and-answer products are useful, but they mostly wait. They sit there until you remember to ask.
That helps when you already know what you need.
It’s useless when the whole problem is that you forgot.
A couple months back I wrote that software has a marginal cost again.
Every AI request has real dollars behind it, and the math gets ugly the moment you scale usage. That matters, because the obvious version of "AI that watches everything" is wildly expensive and probably not very good.
The trick: do not spend frontier (expensive) tokens on everything.
You do not need a genius model to count days since the last reply. You do not need deep reasoning to know a meeting happened, a commitment was captured, or a deadline crossed a threshold.
A lot of the watching layer can be cheap, boring, deterministic plumbing.
Which is exactly what you want.
Then, when something changes (when a goal turns stale, when a customer tone shifts, when a critical reply lands) that is when you spend for judgment.
The watching should be cheap.
Judgment should be excellent.
Get that balance right and the product starts to feel less chatbot and more quiet teammate.
The magic is in the architecture, but good products always hide the tricky bits.
An iPhone just works. You do not think about the tech underneath every tap. ChatGPT gives you an answer in seconds and the math behind it all disappears.
Any worthwhile product has to do the same thing. Especially if the underlying system is weird, messy, and complicated.
What I'm building aims for that shape.
Set a persistent goal and it stays with you until it is closed. Not a chat session. Not a disposable prompt. A real object that can be watched over time.
Email, meetings, chat, calendar events, notes. They all become signal.
Silence gets measured against the horizon you define. Follow-ups get tracked. Commitments can turn into goals. And you decide when something is actually resolved versus merely quiet.
The current version works. The full loop is still being built. Automatic resolution is improving. Meeting-to-goal creation is getting better. Agent outcomes are further out (these things run 24/7 - I also lose track of the work output of each one and have to remember if it was completed).
But the point is not to make the system feel magical for its own sake.
The point is to remove mindshare from work that should not require you to keep a giant open tab in your brain all day.
I've also been running a pilot with a services business, which has been useful because services businesses are basically coordination machines with invoices attached.
Multiple jobs at once. Different owners. Long-lead items. Customer emails. Vendor handoffs. Internal follow-ups.
A dozen threads where "still on track" and "quietly slipping" can look almost identical until it is too late.
And most operators cannot sit in their inbox all day.
They have calls. Jobs. People. Fires. Actual work away from the screen. So by the time they plug back in, the inbox is a pileup, and now the pileup itself becomes the work.
The product is simple from their side: plug in a few jobs and it watches in the background.
Has the status changed recently?
Who owns the next step?
Did the vendor reply?
Did the customer go quiet?
Has the project silently slipped?
Last week the pilot flagged an issue that had gone quiet for a few weeks. The team had not noticed yet. It probably would have sat there for several more days, maybe longer, because nothing was obviously on fire.
That is the worst kind of miss.
Not dramatic nor loud…just absent.
The product caught it and escalated.
The operator did not open an AI chat and ask, "What needs my attention this morning?"
He just looked at what had been flagged and acted.
That difference matters.
People are busy. They do not want another place to search.
They want to know what deserves attention and skip over the rest.
That is why there are so many AI inbox tools, meeting tools, and assistant tools right now. The pain is obvious. The solve is still early.
Most of these products live in isolation. They summarize one surface. They rewrite one email. They answer one question.
Useful, but incomplete.
More interesting products understand the workflow enough to know when reality has drifted from the plan:
Flag any deal where the champion has gone silent for a week.
Let me know when a follow-up I promised gets missed.
Tell me when the customer tone shifts.
Watch this renewal until the paperwork is done.
Escalate if this long-lead item goes quiet.
For operators, that looks like a workflow product.
For builders, it looks like infrastructure: goals, commitments, silence events, follow-up events, status changes. Simple primitives they can wire into their own agents and products.
Same engine, two surfaces.
Last year everyone was chasing the chatbot on top of company data.
Query. Answer. Repeat.
That was useful, and still is. But it was never the end state. A text box is only helpful when you remember to use it.
Better still, focus on what AI can notice before you ask:
What should it carry for you?
What should it watch?
What should it escalate?
What should it keep out of your head until it matters?
A month ago I had a conversation I wanted to build on. A few weeks later I had completely lost it in the blur.
Luckily, my product remembered when I didn't.
That is not everything, not even close.
But it is the right shape.
Not answers. Attention.
Join the Fin and AWS team on July 9 to see how you can scale customer support with AI. Save your spot.
