I am not using AI like a chatbot anymore

WEBVTT
00:00:12.129 --> 00:00:14.769
Most people are still using AI like a search box.
00:00:14.849 --> 00:00:20.530
They open a tab, ask a question, copy the answer, paste it somewhere else, then the context disappears.
00:00:20.769 --> 00:00:25.570
Useful, yes, but it is also the least interesting version of what this technology can become.
00:00:25.729 --> 00:00:47.329
The real shift happens when AI stops being a place I visit and starts becoming a layer across my work, memory, tools, documents, scheduled research, messaging, search across past conversations, reusable workflows, a system that helps me stop losing the thread.
00:00:47.649 --> 00:00:50.849
That is what I have been building with Frank and Henry.
00:00:51.009 --> 00:00:53.969
Henry runs on OpenClaw, Frank runs on Hermes.
00:00:54.210 --> 00:01:01.729
They sit on two small Mac systems in my home lab, not on the district network, and not as a district managed AI system.
00:01:01.890 --> 00:01:02.609
That matters.
00:01:02.769 --> 00:01:03.890
This is my environment.
00:01:04.049 --> 00:01:12.769
I am experimenting so I can understand the workflow, the risks, and the design patterns before making claims about what any of this should mean for schools.
00:01:12.930 --> 00:01:14.370
Henry is my thought partner.
00:01:14.450 --> 00:01:23.250
He is where I test ideas, push on framing, work through research questions, and let messy thinking get more organized before it turns into something public.
00:01:23.409 --> 00:01:24.689
Frank is my orchestrator.
00:01:24.769 --> 00:01:34.450
He helps preserve context, monitor signals, manage scheduled research, search across memory, work through documents, and connect different parts of the system.
00:01:34.609 --> 00:01:35.649
Henry helps me think.
00:01:35.810 --> 00:01:36.849
Frank helps me move.
00:01:37.009 --> 00:01:39.810
Together, they point to something bigger than a better chatbot.
00:01:39.969 --> 00:01:42.370
A chatbot answers the prompt in front of it.
00:01:42.530 --> 00:01:45.250
An operating layer preserves context.
00:01:45.489 --> 00:01:49.329
Roots work schedules attention and reduces the cost of starting over.
00:01:49.489 --> 00:01:53.409
Most knowledge work does not fail because people are not smart enough.
00:01:53.569 --> 00:01:55.409
It fails because context leaks.
00:01:55.649 --> 00:01:57.489
The meeting happened, but the follow-up died.
00:01:57.649 --> 00:02:00.929
The article was useful, but the idea never made it into the system.
00:02:01.009 --> 00:02:02.929
The email mattered, but it got buried.
00:02:03.090 --> 00:02:09.250
The same report gets rebuilt every month because no one can find the last version.
00:02:09.409 --> 00:02:11.250
That is the problem I am trying to solve.
00:02:11.409 --> 00:02:14.129
Not how do I make AI write more words?
00:02:14.370 --> 00:02:15.890
We have plenty of words.
00:02:16.129 --> 00:02:17.890
The internet is choking on them.
00:02:18.050 --> 00:02:27.250
The better question is this: how do I build a system that helps me notice what matters, remember what matters, and act on what matters?
00:02:27.409 --> 00:02:29.009
The model is not the system.
00:02:29.169 --> 00:02:31.729
The workflow around the model is the system.
00:02:31.969 --> 00:02:37.729
If the output dies in the chat window, I do not have an operating layer, I have a very articulate scratch pad.
00:02:37.889 --> 00:02:50.210
The system starts to matter when the assistant can remember context, load workflows, search prior conversations, work through documents, run scheduled research, and ask for approval before taking external actions.
00:02:50.289 --> 00:02:51.409
That last part matters.
00:02:51.569 --> 00:02:53.250
External actions need approval.
00:02:53.490 --> 00:02:55.409
Drafts are safe.
00:02:55.810 --> 00:02:59.009
Private context stays private.
00:02:59.250 --> 00:03:06.609
Publishing, posting, replying, sending, deleting, or sharing require a human decision.
00:03:06.770 --> 00:03:08.530
That is not distrust.
00:03:08.770 --> 00:03:10.930
That is trust with architecture.
00:03:11.090 --> 00:03:14.129
The most practical part of this setup is scheduled research.
00:03:14.289 --> 00:03:21.969
Frank can run recurring jobs that bring me briefings, trends, writing ideas, and weak signals before I remember to go look for them.
00:03:22.129 --> 00:03:24.050
Not everything needs an AI model.
00:03:24.129 --> 00:03:27.490
Sometimes the right answer is a script, sometimes it is an agent, sometimes it is both.
00:03:27.650 --> 00:03:29.250
This connects directly to education.
00:03:29.409 --> 00:03:33.409
Most AI conversations in schools still sound like tool conversations.
00:03:33.569 --> 00:03:35.090
Which chatbot should we use?
00:03:35.250 --> 00:03:36.449
What should the policy say?
00:03:36.609 --> 00:03:37.729
How do we stop cheating?
00:03:37.889 --> 00:03:39.409
What prompt should teachers try?
00:03:39.569 --> 00:03:40.449
Those are real questions.
00:03:40.530 --> 00:03:41.250
They are not enough.
00:03:41.409 --> 00:03:43.329
Agents push on the structure of work itself.
00:03:43.490 --> 00:03:55.889
If an AI assistant can remember, search, schedule, draft, verify, use tools, and work through documents, the question changes.
00:03:56.129 --> 00:03:58.050
Where does context get lost?
00:03:58.210 --> 00:04:01.490
Which workflows depend on one person remembering everything?
00:04:01.650 --> 00:04:04.609
Which decisions stall because the information is scattered?
00:04:04.770 --> 00:04:09.810
That is why I keep coming back to intentional design.
00:04:10.129 --> 00:04:15.650
AI should not be bolted onto broken workflows and called transformation.
00:04:15.810 --> 00:04:19.810
That is just digitizing compliance with better auto-complete.
00:04:19.970 --> 00:04:34.370
The opportunity is redesigning the work, not replacing human judgment, not outsourcing relationships, not automating responsibility, redesigning the system so people spend less time reconstructing context and more time making better decisions.
00:04:34.529 --> 00:04:36.850
Frank and Henry are not replacing my thinking.
00:04:37.009 --> 00:04:39.490
They are helping me stop losing the thread.
00:04:39.569 --> 00:04:41.250
And that is a much better goal.
00:04:41.410 --> 00:04:43.969
The point should not be to make everything more automated.
00:04:44.129 --> 00:04:47.009
The point should be to make the work more intentional.