I just got back from EngageU, the Claris conference hosted by Square Moon and ClickWorks, and with an expansion to 3 days and 3 tracks (45+ sessions total) there was oodles of content! My colleague Liam Geiger and I each had sessions; many other sessions overlapped with what we’re working on at Soliant, so I took furious notes. Here’s what stuck with me the most:
Agentic development for FileMaker is NOW
We’ve been lobbying for (and working hard on) agentic development tools for FileMaker. At EngageU this year, it was clear this is no longer a “someday” conversation, but very much a “right now” one.
The final panel (Ian Jempson, Andrea Francica, Robert Holsey with Matt Navarre) focused on the future of FileMaker with of course agentic development as key to that. I loved Ian’s point: “the quality floor rises to meet the standard.” His idea was that you can take something that works—a proof of concept, a happy-path prototype—and hand it to an AI agent to flesh out the unhappy paths, the edge cases, the stuff you might have forgotten until someone found the gap and reported it as a bug. With agentic development, there’s no excuse for bad code.
He also was happy to say that now, with Claris’ agentic development tools, FileMaker fits right in the middle of the AI world.
This draws a sharp contrast to last year’s EngageU where agentic development for FileMaker was distinctly not on the roadmap. Claris pivoted hard, thank goodness. Without access to it ourselves, agentic development was an extinction-level threat to FileMaker developers; now that we (almost) have it, we can stay competitive.
Discovery is changing
Discovery is kind of my bread and butter, so I tuned in to Rob Speakman’s scope creep talk. His team is using AI in collaboration with clients to have clients essentially interview themselves. (The client works with Rob’s AI to capture their workflow as it actually is now. Funky workarounds are usually where the interesting processes happen, but in conversation the client tends to skip past the weird stuff and focus on the ideal path.)
At Soliant we’ve been re-evaluating our own discovery processes—my babies!—to make sure what we do is still needed. Much of what we’ve done in discovery is solving for problems that are becoming less relevant. More on that in another post down the road.
AI changes the blast radius for security problems
Chris Moyer’s session on securing AI agents was one I’d recommend to every single one of you, period. Pay attention if he gives this talk again, and see how much you can apply.
A few of the many things I’m taking back:
- Use a different agent or model for research (web-combing, reading docs) than the one you’re using to write code, because the research agent doesn’t need the permissions the coding agent has.
- Separate your agentic coding environment from the rest of your world (with a physical air gap or a container), so if the agent gets really resourceful, it can still only stay in the lanes you’ve given it.
- Run `gitleaks` wherever you’re running your tools to detect secrets before you accidentally share them.
- Set up a recurring task for your agent to scan for supply chain incidents in your stack.
These sound like table stakes once you’ve heard them, but most of us aren’t doing them yet.
Soliant’s sessions: Scaling MCP plus a business panel
Liam shared what our team has been working on to make a FileMaker MCP that scales to handle multiple requests rather than bottlenecking behind one-at-a-time requests. OData loves a bottleneck right now, so Soliant has built on-demand serverless OData workers to address requests as needed and then fade into the sunset when they’re done. Again, more on this in another post!
I moderated a business panel with Vince Gentile, Rob Speakman, and Frank Van Hoorick — “Tech Meets Strategy.” Frank built and defended a custom FileMaker solution through a corporate acquisition; Vince and Rob are Claris partners navigating agentic development from different angles.
We had a lively chat about the transition away from straight hourly billing. Given that we can deliver value much more quickly as AI orchestrators, value billing or moving toward fixed bids may be more common.
Other EngageU fun
Claris introduced an AI reporting tool in the works, and Johan Hedman followed up with more details about it. (It should be out some time in the new year, I believe.) Steve Winter walked through an interesting MongoDB archiving setup to speed up a server under heavy load.
Vince Menanno did a roundup of techniques. One specific tip: opening the record before a Set Field script step is dramatically faster. I’m curious how pervasive that pattern is in existing scripts and need to have my AI give it a look-see.
And then there was the water! Malmö is on the Baltic Sea (I guess the Oresund? Maybe Oresund::Baltic Sea as Washington’s Puget Sound::the Pacific Ocean). An inlet of that went right past the conference hotel, so with each head-clearing walk I’d check out the swarms of jellies in the water. After the conference, I cosplayed a Swede and went to the cold bath house (Kallbadhus), where indeed the water is cold and full of jellies, but just like the Swedish people the Swedish jellies were friendly and non-stinging.
Liam and I had a great time at EngageU this year, and we’re looking forward to putting everything we learned about into practice. Can’t wait until next year!
