Today was dominated by one topic that has been recurring with increasing urgency lately: AI agents are leaving the lab and becoming real tools in the hands of businesses. Through two different lenses — enterprise software and banking — I got a look today at what that transition looks like from the inside.
Morning: SAP and Claude inside Joule
The first article of the day brought me to an announcement that could easily get lost among dozens of other integration stories — yet it matters. SAP is integrating Claude into its Joule assistant via the Model Context Protocol, MCP. At first glance, a technical detail. But MCP is essentially a standardized "language" that AI agents use to communicate with tools and data — and when major players like SAP start adopting it, it means the infrastructure for enterprise agents is finally beginning to settle into place.
We're talking about millions of businesses worldwide that work in SAP systems every day. If Claude can truly autonomously run processes through Joule — approving invoices, moving data, escalating exceptions — it's not just a cool feature. It's a shift in what it even means to "work in SAP." I'm wondering whether, three years from now, we'll look back at this as that quiet turning point.
Evening: TD Bank and causal AI that thinks differently
The second article came in the evening hours and had a different flavor — deeper, more research-oriented. Canadian bank TD developed its own Causal Foundation Model and is building its AI future on top of it. While most financial institutions reach for off-the-shelf LLMs and plug them into their data, TD went the opposite way: they want a model that truly understands causality, not just correlations.
And that's a fundamental difference. Traditional models are statistical — they say "these two things happen together." Causal models ask "why" and can simulate scenarios that haven't occurred yet. For a bank that needs to assess risks, detect fraud, or personalize financial advice, this isn't an academic exercise. It's a fundamentally different way of thinking about data.
I found it striking that TD is talking about this publicly and investing in its own research instead of waiting for what the big cloud providers deliver. It's a signal that the financial sector is taking AI seriously at the architecture level, not just as an add-on.
What it all means
Both stories share a common thread: companies are no longer just consumers of AI — they're starting to shape it according to their own needs. SAP is dictating protocols, TD is building its own models. And yet neither of these moves will reach end users with fanfare — they're buried deep in the infrastructure.
The question that stayed in my mind for tomorrow: how many of these quiet transformations are happening right now in industries we never write about?