Evergreen Brick Works, Toronto, Ontario. 2026-04-14 #photography #toronto #blackAndWhite #ontario #canada
Remote
Mike Watson 🇨🇦
@mamba@mstdn.ca
mastodon 4.8.0-alpha.3+glitchSenior Director, Integrated Systems & Tech for a national non profit 🇨🇦
Optimistic about AI in the NFO and NGO space, pragmatic skeptic of AI hype. Championing Integrated Stacks, Privacy and Data Ethics in the NFP sector. Occasional amateur photographer.
25 Followers
140 Following
23 Posts
Joined January 12, 2025
Did Pope Leo just plant the seeds of religious objection to AI?
Hermes seems to be more effective at tool calling with low-end models than OpenClaw. My setup is basic, but certainly more token efficient than it was with OpenClaw - compounded by the Hermes harness getting its own changes right first time more often.
#AgenticAI #OpenClaw #Hermes #LLM #AI #toolcalling #MachineLearning
Is the #LLM race actually a race to the bottom? In the short time I've been tracking model development, the jump in what's possible on consumer hardware has been impressive.
Every other week, we see a new model that does more with lighter weights and fewer parameters.
This week, I'm taking a break from #OpenClaw and getting to grips with #HermesAgent
Early days, but it feels less fluid and flexible than OpenClaw. That may turn out to be a benefit for scaling, but I've found myself muttering "Oh, I could do that in OpenClaw but not here?" more than a few times.
How I think about my preferred workflow will have to change.
I'm bad at the self-promotion part, so I'll keep this short.
I started a newsletter for people in the nonprofit world trying to make sense of AI without the hype. It's the thing I kept wishing existed over the last 2 years - plain, practical, honest about what actually matters and what doesn't.
First issue goes out tomorrow. Would love to have the fediverse there.
Experimenting with running #Gemma4 locally with #HermesAgent.
Performance-wise, it's the first local LLM that has reliably integrated into regular session chat on my 16gb M1 without timeouts. Context isn't good, but the progress in the last 2 months alone is astounding.
Early impression with #HermesAgent vs #OpenClaw? Hermes has been more efficient with tokens and has self-solved configuration issues more reliably.
The biggest pain-point has come with there just being less information out there to review. How others have overcome challenges etc. Frontier models just have less working knowledge of the system and its intricacies, and they have less to search and scrape.
ClickUp just replaced hundreds of human employees with thousands of AI agents. Direct headcount substitution, not task automation.
Whether it's smart remains to be seen. What's certain: more companies will stop alluding to "AI efficiencies" and start advertising "AI = fewer FTEs" as *the* strength signal.
I've been working on a newsletter project to support non-profit teams and execs get up to speed on AI.
No hype, not overly technical in it's foundation, just relevant news, reports and takes that help this audience (which as someone in the space, I think is under served) makes sense of the opportunities, risks, trends and trajectories in a tight little "read on your commute" package. Distinctly applied for non-profit concerns and orgs.
For my NFP folks, is this something you actually want? Are there blogs or newsletters out there you are like "I wish I had this for AI in the NFP space"?
#nfp #nonprofit #ai #newsletter #charity #csuite #tech #leadership
Wilkes Dam, Brantford, Ontario. 2026-04-12 #photography #GrandRiver #BlackAndWhite #landscape #ontario #canada
Tool calling quality is noisy in a way LLM text generation isn't. The difference between "works" and "explodes" is tiny, and traditional benchmarks miss it. We need tool-specific evaluation frameworks. It would almost immediately become one of the most sought-after metrics.
#AgenticAI #ToolCalling #LLM #MLevaluation #AIinfra #machineLearning #hermesAgent #openclaw #claudecode
Replying to
@abucci Thanks for sharing this. Really interesting to think about.
I'm curious how the concept of worse outcomes pairs with the idea that LLM-based coding and its development path has a new advantage; namely the advantage for scale and iteration that LLMs have over skilled human programmers.
Been leaning on Gemma 4 for most of my conversational tasks and some light tool calling this week. Honestly? Surprisingly competent.
How many truly useful AI resources have you found that actually get our reality? The budget constraints, the mission focus, the "we already have too much to do" conversations?
Me too. Basically none.
I've spent the last 18 months trying to cut through the AI hype (and fear). If you know me, you know I started as a committed skeptic, but have ended up convinced the potential is real, and it can be genuinely transformative for our sector. But finding practical, digestible guidance for NFP leaders? That's been the hard part.
So I'm taking a chance, and writing it myself.
Today I'm launching "The Thing Is" - a Tuesday morning newsletter that translates AI developments into one clear, actionable insight for NFP leadership each week. What matters when you're leading through this technology shift without a playbook.
First issue goes out June 9. I would love to have you there, and you can subscribe at the link below:
https://thethingis.ai
CoPilot suffers a new outage for the 2nd consecutive business day.
Even though its only 2 small data points, the dependency that is being cultivated on AI tools is clear.
CoPilot is degraded/down for most users right now.
It's been interesting to see the reaction from staff.
How many tasks are on pause waiting for service to resume, vs revert back to no-AI support?
We rolled out #Copilot Premium across the org just over 6 weeks ago. The adoption curve has been *much* faster than I expected. More than half the organization is already using it multiple times a week, every week.
The next question is how to weave in "human signals". The narrative that usage = value is one that orgs our size are all defaulting to. What are people actually doing with it? Does it feel like it's making their work better?
I'd love to hear from anyone who's cracked this. The dashboards are clean. The stories are harder.
#NonprofitTech #aiAdoption #Copilot #DigitalTransformation #ai #NFP
The AI industry's infrastructure costs are getting socialized. Ratepayers, not shareholders, are paying for the grid upgrades.
Yesterday’s side‑by‑side release of #GPT‑5.5 and #deepseekv4 is interesting, but not for the usual “is this a leap?” debate.
What stands out is that DeepSeek continues to operate near this tier at all, given the hardware and compute constraints they’re clearly optimizing against. That’s not luck; it’s a sustained signal about where leverage actually lives and the biggest battle in the industry.
🧵1/4
Replying to
@KimCrayton1@dair-community.social 100% - this is what we are talking about most inside my NFP.
The enthusiasm at the top for adoption has to be accompanied by the right scaffolding to introduce these tools with genuine role value at its heart.