What we're learning from building private, governed AI for real teams — what works, what doesn't, and what to sort out before you start.
Models are a commodity. The harness that holds your AI's memory, skills and credentials is what matters, and it's the part to keep under your own control.
Read the post →How an AI bot in Microsoft Teams now finds, fixes and opens a pull request for regression bugs, while a person still reviews every change and owns deploys.
Read the post →Windows PCs can now run local agents. That's the personal layer. Expert bots with logged, role-scoped workflows are the layer that helps the business run.
Read the post →Autopilot lives in Microsoft 365, but many businesses still run Sage 300, QuickBooks Desktop and homegrown apps. Who handles everything Microsoft can't see?
Read the post →OpenClaw Enterprise builds its story around running agents on your own infrastructure. Why that matters, and the gap for businesses without engineers to run it.
Read the post →Automating onboarding and offboarding sounds easy, but every automation needs somewhere to run. How AI agents give small-business automations a home.
Read the post →Governance has to live where the data lives. Why Toodlebox builds AI on hardware customers control, and why the biggest players are moving the same way.
Read the post →GenBio AI's AIDO Cell simulates a human cell digitally, so drug ideas can be tested faster and cheaper. A reminder that not everything AI touches is scary.
Read the post →Gossip protocols predate Bitcoin. How Toodlebox is using crypto's useful idea, nodes syncing with no single point of failure, without paying for trustlessness.
Read the post →Toodlebox is moving its backend from Cassandra to PostgreSQL to free up memory for AI agents running on small, local hardware. Here's the reasoning.
Read the post →One bug led to years of hidden technical debt. With AI helping untangle it, a half-finished refactor finally shipped. Not replacing the developer, amplifying one.
Read the post →AI can write code faster than you can understand it. Why asking for small, verifiable chunks is what gives you the confidence to press deploy.
Read the post →A plain-English request over Microsoft Teams replaced what would have been an hour or more of developer work: changing an AI agent's alert schedule.
Read the post →94% of MSPs are investing in AI governance, but only 43% say they can deliver it. Who catches it when an AI tool does something nobody authorized?
Read the post →Only about 5% of enterprise Copilot pilots expand into a full rollout. Turning Copilot on isn't a strategy without usage governance and someone reviewing the output.
Read the post →A popular project management tool that never fit a client's team taught us to start with the shape of the business, not the tool everyone else uses.
Read the post →Software that waits for someone to log in can't keep up with AI that does the work. Why Toodlebox is rebuilding around governed AI in your own environment.
Read the post →