Know what every AI agent is worth: beyond token counting

Here's a question for your next leadership meeting: which of your AI agents would you retire tomorrow?

Every enterprise has candidates. It might be the agent built for a pilot that wrapped months ago. The one with a dozen users and a token bill that keeps climbing. The one your team swears by and yet nobody has measured.

You're paying for all of them, but it’s probably unclear which ones are paying you back. 

That question is heading straight for the CIO's desk. It's coming from the CFO, the board, and every business unit asking for more budget to build the next AI agent. According to IBM, only about 25% of AI initiatives have delivered their expected ROI.

Agent spend is climbing, and budget reviews are getting harder. The enterprises pulling ahead can say exactly what each agent is worth. Here's how enterprise leaders put cost, risk, and business value side by side for every agent, and use that evidence to decide what to scale, refocus, or retire.

The courage to kill projects

At Dataiku Succeed 2026, Dataiku CEO Florian Douetteau drew a sharp line between two kinds of enterprises. Company A fears governance will slow it down. Company B sees value and risk together, and uses that visibility to optimize. Florian argued that every enterprise can buy the same models but the winners bring strategy, implementation, and the courage to kill projects.

In most enterprises today, agents get funded by whoever tells the best story in the budget meeting. Meanwhile, Company B funds them on evidence.

A price tag and a scorecard for every agent

We built Dataiku Agent Management to put cost, risk, and business value side by side for every agent running across your platforms. Each one is measured on business outcomes as well as technical performance.

• Cost. Token consumption per agent becomes a cost figure, tracked month over month next to real usage from actual traces.

• Value. You define what success looks like for each agent, such as escalations avoided or hours saved. Agent Management supplies the pipeline and the display, so you can see the agent’s contribution in business terms, right next to its cost.

• Risk. Each agent's risk is assessed and kept current, so you can see where exposure concentrates across the portfolio.

Side by side, those numbers tell stories that usage alone would hide. Your busiest agent might be burning tokens on questions it was never built for. A quiet one with a few dozen users might be saving a team hours every week. "AI is expensive" becomes "these three agents earn their cost, and this one serves eleven people."

The CIO demo at Succeed showed exactly that picture: average cost per agent, token allocation, cost by month, and business KPIs attached to each agent. A global risk view showed the governance team where to focus.

Beyond tokens and counts

At Dataiku Succeed, Levi Bowman, Sr. Director, Data, Analytics & AI, summed up Solidigm's philosophy as "AI for everyone, by everyone." The people who own the outcomes build the agents. So Solidigm starts with an inventory, then measures each agent's effectiveness against business outcomes, well beyond tokens and counts.

That's what makes "by everyone" sustainable. When every agent's contribution is on the record, you can let more people build with confidence.

Start while it's still an inventory

Managing the complexity of everything enterprises have built is shaping up to be the defining challenge of the next six months. Supervision only gets harder from here. Your agent count is growing whether anyone is watching or not. Start now, and your portfolio stays an inventory. Wait, and it becomes an investigation.

Agent Management keeps the evidence, and your teams make the call. Scale the agents that deliver, refocus the ones that drift, and retire the rest.

Every agent has a cost. Company B knows which ones are worth it.