One Engineer, $4,000 in Claude Credits: Why Paessler Now Says No to AI Spend

By Priya Raman

An engineer blew through $4,000 of Claude credits by the 17th of the month, then asked for more. That request, Paessler says, is exactly why the company now treats AI spend like any other line item — capped, reviewed and, when necessary, refused.

Alerts had already fired at 75% of the monthly allocation when the ask landed. Paessler's reaction was blunt: would anyone approve conventional IT at a few thousand dollars per seat, per month, with fluctuating costs and no guaranteed ROI? Probably not. The German network-monitoring vendor decided the answer for AI shouldn't be different.

The wider numbers back them up. McKinsey reports 93% of companies exceed their AI budgets. A CIO in Paessler's network watched an annual AI budget evaporate within two months. Gartner predicts AI coding costs will surpass the average developer's salary by 2028. The tools may soon cost more than the person using them.

How did the bills get so big?

Paessler's first AI-focused dev day set the pattern. The company offered a $5,000 prize for the best feature built into its monitoring platform and got a record 29 submissions — context-aware assistants, root cause analysis helpers, some deliberately using the most expensive approaches available. Paessler encouraged it and gave cost "a very long leash."

Then the invoices arrived. More prompts meant more tokens; multi-step agents burned credits faster; newer models and high effort levels sent bills, in the company's words, "to the stratosphere." AI's one-two punch — consumption-based pricing plus employee curiosity — is a dangerous combination for enterprise budgets, especially as major providers start metering heavy, agent-style usage separately.

Is a spending cap a ban?

Paessler insists it isn't. The company's playbook:

• Set group limits and token ceilings at rollout, not after — you can't retrofit discipline onto a tool people already use freely.
• Work backward from what you'll tolerate spending per person or team.
• Raise the ceiling when value shows up.

Caps, the company argues, convert invisible outlay into deliberate decisions. An executive hitting a monthly cap during heavy pre-board-meeting analysis gets a quick check and an increase. A user tinkering with no critical business need gets a manager's pause. Same cap, two very different outcomes. Both work.

Employees aren't doing anything wrong, Paessler stresses — they're legitimately experimenting, usually with positive results. Too much of a good thing simply hits the bottom line. Culture, not just controls, keeps them cost-aware.

What about shadow AI?

Two-thirds of office professionals at large companies admit using AI tools at work they believed weren't permitted, according to PagerDuty research. That's waste — paying for capability you already own — plus unmanaged attack surface. Consolidation does double duty: cutting spend and shrinking exposure.

Paessler's answer is a cross-functional AI Center of Excellence, meeting weekly, with decision makers from IT, information security, HR, finance, legal and operational excellence. Before approving any new tool, the CoE asks whether something already in the estate can do the job through an integration. Its pen tester then tries to break each new AI app by accessing data he shouldn't see.

Education matters too. Claude, for instance, lets users pick the "effort" level for a task; higher effort means more tokens and higher cost. Training employees to match model selection to actual need, rather than a default setting, prevents overruns.

The CoE has become Paessler's single voice on guidance, approved use cases and training, with one tool repository and one set of usage standards — useful for compliance, since the company always knows what's running and why.

And sometimes the answer is no. One app jumped unexpectedly in cost; Paessler discovered its main AI platform could handle the job anyway, and consolidated. As CIO's Jay Ferro wrote in September, many companies got swept up in the hype and forgot that AI is a tool requiring a business case.

Paessler's verdict: tokens for the sake of tokens isn't sustainable. In the year ahead, "no" should be the most valuable word in your AI vocabulary — not because AI lacks value, but because constraint is how you maximize it.