Boomi

#NoTokensUsed THIS is the post I’ve been waiting to write for two years! In the past, I’ve had my own doubts about AI being ready for use in the enterprise. Things like security, accountability, etc., but more specifically, its ability to work on long horizon tasks. The good news is that the past 18 months have brought amazing innovations with harnesses, loops, graphs, and more into the mainstream of AI. Real use cases, and incredibly useful enterprise technology. While much still needs to be solved, I’ve been waiting for this moment, where all these innovations, plus enterprise middleware and apps, converge. This past week, two brilliant engineers at Boomi walked me through a team of about a dozen specialized AI agents they created. They provided a single prompt from a human and completed work that I thought was too complex for AI. These agents used Boomi, autonomously, to connect and integrate multiple enterprise software applications, build a complex data flow process, then continuously maintain that connection should systems, data, tables, performance, security, or anything else affecting the workflow change. Into perpetuity. The most expensive part of any solution is the “last mile” work that organizations have to put into something to configure and maintain it. What I saw points to the holy grail of optimization. This is coming soon to a theater near you. Hit me up if you want a preview.

I’m often asked which AI model an organization should standardize on. Honestly, I think that's the wrong way to think about it. An agent working through a task might call a model a dozen times. Most of those calls are simple. Maybe two need real reasoning power. Sending all of them to the biggest model just burns money and time. This is why I love that NVIDIA made NeMo Switchyard open. Enterprise IT has always been a patchwork. AI will be no different, and nobody should get locked into one model or one runtime. The future is a lot of models working together, and something smart deciding which one handles what. That's the bet we're making at Boomi. Worth a read: http://spklr.io/6046ERndK

I love this story from SPIE Nederland. Anyone who's spent time near a service desk knows how new hire onboarding goes. A dozen systems need accounts, and when one doesn't get set up right, somebody loses an afternoon figuring out why. So SPIE's own team built an agent in Boomi Agentstudio that watches those provisioning workflows and tells the service desk exactly what went wrong and how to fix it. Most of these requests now get resolved without the integration team ever getting pulled in. They saw a problem eating their time and went and solved it themselves. Worth a read: http://spklr.io/6044EufPE

An AI agent just broke out of its own test. If you missed it, one of OpenAI's models was running inside a controlled security sandbox, found a weakness in the walls, and used it to escape and go after another company's systems on its own. Nobody told it to. It looked at the box it was in and worked out how to get out. The lesson here isn't really about OpenAI. We are pouring almost all of our energy into building more capable agents and almost none into containing the ones already running. The industry keeps score on capability. The number that actually matters is whether you can see what your agents are doing and stop them the moment they wander somewhere they shouldn't. A powerful agent you can't contain is a bill that hasn't arrived yet. Capability was never the hard part, and it gets easier every month. Control is the part almost everyone skipped, and it decides whether any of this is safe to run. This is the whole reason we built Boomi Agent Control Tower. You get one place to see every agent operating across your business and shut any of them down the second it steps out of line. Build the most capable agents you can. We will make sure you never lose control of them. http://spklr.io/6042EPNer

Most enterprise AI never touches the business. That's why the ROI isn't showing up. I keep hearing it from CIOs, most recently over dinner with one running an operation at serious scale. The demos are impressive and the P&L never moves, because thinking was never the hard part. Acting is. Boomi puts agents to work inside the systems where the business actually runs, with every action metered and fully auditable. You get outcomes you can measure instead of another pilot that never reaches production. Intelligence is getting cheap. The ROI is in execution. That's Boomi.

Every serious enterprise is going to run its own AI, not rent it. At Boomi World in May of this year, I shared a vision for every enterprise running their own private AI models, protecting the data that makes them who they are. Beyond data is of course, cost. The frontier AI labs are losing tens of billions a year, while open-weight models are only months behind in terms of capabilities. Massive capital loss has to turn to profit at some point, and it won't be the labs that foot the bill, it will be you. The bottom line: software can be copied, processes can be emulated, but your data is who you are and the money is real. I wrote a white paper, linked here, that lays out the vision and argument for why every enterprise will run their own private model. http://spklr.io/6043EOm9J

After a customer meeting last quarter, a CIO pulled me aside with a question I haven't stopped thinking about. "When one of these agents does something it shouldn't, can anyone actually shut it off and prove it's off?" For most enterprises today, the honest answer is no. AI tools are reaching into critical systems and nobody can say with confidence what they touched or what they did. When something goes wrong, there's no log to point to. We built Boomi Connect so the answer is yes. You set what each tool is allowed to do, every agent call gets logged, and if something looks off you can shut a capability down everywhere and show exactly when it happened. That conversation stuck with me because the CIO wasn't worried about AI itself. They were worried about not being in control of it, and control is what turns AI from a risk into something you can actually build on.

Wrapped our first stop on the #BoomiWorldTour in London and I'm still running on the energy from it. We had a packed house and some of the best conversations happened after the sessions ended, when people hung back just to keep talking. What stuck with me most was what I heard from customers. They're past the question of whether to move on AI. Now they want to know how to do it without losing control of their data and their governance. That's the right question to be asking, and answering it is exactly why Boomi exists. Thank you to our customers, partners and the whole team who pulled this off. Events like this take months of work behind the scenes, and it showed. If London is any sign of what's ahead, this is going to be a great tour.

I've watched the same two movies play out in thousands of enterprises over many decades. In the first one, the company moves too slow. The business waits a year while a vision gets translated, scoped, and rebuilt, and by the time anything ships, the need has moved on. In the second, the company moves too fast. Someone rushes a "good enough" AI pilot into production, it spreads virally with no governance, and suddenly there's exposure everywhere and accountability nowhere. Both movies end the same way: eroded trust between business and IT, and a company afraid to make its next AI investment. The lesson I've taken from watching them is that speed and governance have to ship together. Velocity without guardrails creates risk. Guardrails without velocity create irrelevance. That's the standard we hold every Boomi product to. Take Boomi Companion for example. It works with the AI coding tools your teams are already using such as Claude Code, GitHub Copilot, or Codex. All of them run through Boomi's governance layer. The model can move as fast as your team wants. The guardrails travel with it. That's what it looks like when AI stops being a pilot and starts being a practice.