Six months ago, most finance teams were still asking whether AI was worth exploring. On the latest episode of ExtropyOnAIR, Extropy CEO Ravit Gutman and Mike Jasper sat down to grade the first half of 2026 against their own predictions, and the honest answer is: neither of them saw this coming.
“If you fell asleep on New Year’s Day and you woke up today, you’d have missed so much in the AI market and what’s going on in finance with AI,” Mike said. Here’s what actually happened, what it means for finance teams specifically, and what to watch for before the year is out.
Agents went from demo to default
The single biggest shift of H1 2026 wasn’t a new model, it was AI agents moving from proof-of-concept to background infrastructure.
“We’ve seen all the different finance solutions putting out agents within their platform that run autonomously. They run in the background. It’s not something separate that people have to go put in and monitor, but it’s helping to automate different aspects of the workflow.”
The numbers back it up. Citing the Stanford AI Index, Ravit noted that 78% of global organizations have now deployed AI in at least one business function, up dramatically from the under-6% figure they cited for organizations seeing true value from AI back at the start of the year. Within finance specifically, 52% of financial teams are actively piloting or deploying agentic tools today.
Interestingly, the most mature AI application isn’t a finance task at all. It’s software engineering, at 42% fully deployed.
“Even the banks learned to trust AI with writing code, and they did that before trusting AI to manage money.”
The execution gap is still the real problem
Adoption isn’t the bottleneck anymore. Execution is.
“81% of financial departments have tried to adopt AI in one way or another, but 14% call it a transformation that was part of their strategy and they were successful in it.”
That gap, between trying AI and actually transforming a process with it, is the same lesson Extropy has been advising clients on for years: garbage in, garbage out. Piloting AI on top of a messy, inefficient process doesn’t produce a transformation, it just produces a faster mess.
“You don’t want to automate that inefficient process, and you don’t want to underestimate the complexities of things. There is a strategy behind digitizing and building that true data foundation.”
Within the finance function itself, FP&A has seen the most dramatic AI-driven shift so far, in forecasting and analysis specifically, which is now pulling teams back to make sure the core foundational data (AP and AR) is accurate enough to support it.
Buyers changed how they evaluate AI
Ravit and Mike also pointed to a shift in how organizations are approaching AI purchases and partnerships. Early on, most companies defaulted to building AI in-house. That’s shifted toward buying first, proving value through existing embedded tools, and only then building custom solutions for the areas no off-the-shelf agent covers.
“That gave them the experience and that value proof case… And now they’re opening their eyes: we’ve seen so much benefit from these tools that have been embedded in our ERP system, in our order-to-cash system, in our AP system. Now how do we take some of that experience and leverage it in some of the more unique areas of our business?”
Buyers are also more educated and more conservative than they were a year ago. As Ravit put it, CFOs used to just want to hear that a tool “has AI” without asking what it actually does. That’s changing.
The regulation clock is ticking
The AI compliance conversation is catching up fast, and it’s not just a European problem.
“If your AI touches either employment or credit decisions in Europe, it’s very likely that it is your problem. Firms don’t understand that they’re missing those earlier checkpoints and are technically already somewhat out of compliance.”
Key dates on the radar: the EU AI Act’s high-risk obligations hit August 2, and France’s e-invoicing mandate lands in September. Mike drew a direct parallel to how e-invoicing itself evolved from a “wild west” into a heavily regulated process, and argued AI regulation is on the same trajectory, except this time it governs how you get to an output, not just the output itself.
“That’s gonna take a lot of planning, a lot of strategy, a lot of advisory on how do we build something that’s going to meet the regulations as stated, but also make sure everything’s documented… without scrapping and rewriting everything we did.”
There’s also a quieter risk neither regulator has fully addressed yet: concentration risk, the exposure organizations carry from leaning on a small handful of foundation model providers.
Predictions for the second half of 2026
Ravit and Mike closed the episode with predictions for H2, including a rapid-fire round:
- Model choice becomes the boring question. The conversation shifts from “which model” to orchestration, how different tools and agents work together.
- The pilot-to-production gap starts closing, not because the technology improved, but because CFOs are putting AI as a line item in 2027 budgets, with ROI expectations attached.
- The EU AI Act’s high-risk deadline likely slips to 2027, both hosts agreed, citing how overloaded European regulators already are with the e-invoicing mandate.
- At least one major enterprise will publicly blame an AI agent for a material financial error before year end. Both hosts consider this close to a certainty rather than a real prediction.
Three takeaways
- Agents are the infrastructure now. The question every finance team should be asking is which workflow to apply agents to, not whether to use them at all.
- The money follows production, not pilots. Pick one process, end to end, and build a real plan around it rather than running parallel pilots that never convert.
- Governance is no longer optional. Inventory your AI usage before August, regardless of how the EU’s mandate timeline shakes out.
As Mike put it closing the episode: “Six months is a lifetime in the AI market.” Extropy will revisit these predictions on the December episode to see how they held up.
Listen to the full conversation on ExtropyOnAIR.
Book a Discovery Call to talk through where your finance team actually stands on AI, and what a real plan (not a pilot) looks like for the second half of the year.
