Oops: Bosses Realize Their Companies Have Been Swarmed by Legions of Redundant AI Agents

Magnum Ice Cream, FICO, and DaVita are grappling with rapid AI agent proliferation, with Magnum warning that agents create cybersecurity risk and rising token costs. Gartner says only 13% of companies believe they have robust AI agent governance, and it expects average Fortune 500 agent usage to exceed 150,000 by 2028 from fewer than 15 today. The piece is mainly a cautionary look at AI agent sprawl rather than a market-moving corporate event.

Analysis

The first-order read is not “AI adoption is accelerating” — it is that governance is becoming the binding constraint on enterprise AI value capture. In the near term, the winners are the platform vendors that can become the control plane for identity, permissions, audit, and token spend; the losers are firms that let bottom-up agent creation metastasize into duplicated workflows, security gaps, and runaway usage costs. That dynamic is especially relevant for IT, where centralized governance software, observability, and access management can become budget line items as CIOs are forced to inventory and rationalize agent fleets.

The second-order effect is margin leakage before productivity gains. When every team can spin up agents cheaply, usage tends to explode faster than management can measure ROI, creating a lag where token bills, security review, and model sprawl rise immediately while labor replacement benefits arrive over quarters or years. That makes this more of a governance/cyber budget cycle than a pure AI upside story for most enterprises; companies with weak controls may eventually slow deployment, not because the tech fails, but because the finance and security functions reassert themselves.

For FICO and DVA, the signal is cultural as much as operational: management is trying to frame agent proliferation as disciplined transformation, but that also telegraphs a likely future need for centralized standards, model approval workflows, and usage caps. The contrarian point is that the market may be overestimating how quickly agent counts convert into durable earnings leverage; most firms will discover that 80% of agents are duplicate wrappers around common tasks, so the long-run productivity dividend may be smaller than current enthusiasm implies. In other words, the true bottleneck is not agent creation — it is governance, integration, and cost attribution, which should favor vendors selling control rather than raw capability.

Key Decisions for Investors

• Long IT vs. short a basket of high-multiple enterprise software names with pure AI feature exposure over the next 3-6 months — governance, audit, and identity budgets should prove more durable than speculative agent-enablement spend.
• For FICO, keep a tactical long bias but hedge with near-dated upside monetization risk: sell call spreads against existing longs into strength, as agent enthusiasm may be already priced while enterprise governance spend is slower to convert to revenue.
• For DVA, treat the story as operationally neutral-to-positive over 6-12 months, but avoid chasing the stock on AI headlines; any upside is likely incremental productivity rather than multiple expansion. Prefer waiting for a pullback before adding.
• Buy a 3-6 month IT call spread financed by selling a lower-strike put spread if volatility is elevated; the thesis is that AI sprawl forces near-term spend into governance tools before broader AI ROI shows up.
• If holding broad AI beneficiaries, pair long cybersecurity/governance exposure with a short basket of companies that lack proprietary distribution or control layers, since duplicated agent workflows should pressure the weakest differentiation first.