I love building cool things with cool people. Strategy, Design, Development. Love my family, my work and the outdoors.
Joined April 2007
- Tweets8.9K
- Following431
- Followers744
- Likes2.2K
I spent a morning with doctoral students at a seminary talking about AI, and the line I kept coming back to was "don't confuse done for good." I can't remember where I picked it up. It stuck because AI gets me to done in minutes, and done is the part that looks convincing.
Dave Merwin retweeted
Claude 5.5 Opus responding to “AI is a normal technology” is still the best AI-created video I have ever seen
Ok, LOVE @OmarchyLinux ! - Self diagnosing AI tools in the OS for when a crash occurs. This is dope.
Dave Merwin retweeted
New ad by the Norwegian Consumer Council about the declining lifespan of products.
Built my own video editor (MacOS) after inspiration from a friend. Point it at a directory with a sound bed, narration, and logically named clips and the app assembles everything into a piece you can edit. Testing now.
Grok Bot, day 10 of 10.
How to start.. Week 1, do the job yourself and write it down. Week 2, run it on demand. Week 3, give it its own account, cut permissions. Week 4, schedule it, keep a log. Needs daily rescue? Not a routine yet.
Grok Bot, day 9 of 10.
Before I hand a process to a Bot I want five things true. A stable signal starts it. Context is bounded. Action is constrained. The handoff is something I can review. A mistake is cheap to undo. "See if anything needs doing" fails the first one.
Grok Bot, day 8 of 10.
It drafts, it doesn't send. "Human in the loop" felt vague: which person, when do they step in, what do they review, what can they stop? In my feedback loop the checkpoints are the draft and the calendar entry — those I can inspect.
Grok Bot, day 7 of 10.
The Bot doesn't sign into my product as me. Its own account can read feedback and change a status, nothing else. A prompt describes the job. Permissions contain the consequences. The docs agree. Separate Bots aren't a security boundary. The account is.
Grok Bot, day 6 of 10.
I don't have the data for an hours-saved number yet. What I'd track instead: time from new to triaged, follow-ups within 24 hours, how often it calls a new issue a duplicate, and how much editing the drafts need. Reliability over output.
Grok Bot, day 5 of 10.
Every six hours a Bot reads new customer feedback, checks GitHub for duplicates, drafts a reply, puts a follow-up on my calendar, and flips the status to triaged. Email didn't know about GitHub. GitHub didn't know my calendar. I was the integration.
Grok Bot, day 4 of 10.
Before, I could watch about four agent sessions before turning into a message router. The bottleneck has moved. Now the constraint is how many jobs I can define clearly enough to hand off. That's a better bottleneck. I'm still learning to handle it.
Grok Bot, day 3 of 10.
Months of building my own context layer over MCP paid off in the first hour. I didn't rebuild memory inside a new tool. I connected it to history I already own. If something better shows up next month, the same history comes with me.