You spent more on AI this year. Your headcount did not go down.
Harness surveyed 700 FinOps and engineering leaders in July. Roughly one in four dollars spent on AI goes to waste. More than half said nobody in their company owns AI costs.
The Domino Enterprise AI Report, also released in July, surveyed 639 senior AI leaders. 57% of enterprises see returns that fail to outpace what they spend.
I speak with senior executives every week about what they are doing with AI. The ones producing results have stopped asking what to add. They ask what to remove.
Here are five things they are removing right now.
1. Remove the Headcount Assumption
Most executives ask AI to help their teams move faster. That question produces small gains. A team told to speed up will optimize the work it already does. A harder question: what would this team deliver if I took the resources away?
Keith Zubchevich, CEO of Conviva, declared his company an AI-first business. Every function and every person had to treat AI as part of the job. Then he put his leadership team in a corner.
“You get two choices… You either cut your team in half, halve the number of people, or you give me 5X your commitment from last year.” – Keith Zubchevich, CEO of Conviva
His leaders had to redesign the work. The organization came out with 3X the productivity.
Put the same question to your leadership team this quarter. What will they ship with half the people?
2. Remove the Ceiling on Output
For thirty years, output scaled with hiring. Want more code, hire more engineers.
Jascha Kaykas-Wolff, CEO of Visiting Media, broke that link himself. He built a three-tiered operating model run by agents, including ones he named Mira and Nova. They write code, review pull requests, and work beside his product managers as team members.
“I have personally… built out… a relatively interesting fleet of agents that I’ve been working with for the last about nine months…” – Jascha Kaykas-Wolff, CEO of Visiting Media
Read the first two words again. He built it. Nine months of daily practice, by the chief executive. He argues that leaders who delegate this learning will never bring anything useful back to their teams. The useful parts show up only in the hands-on work.
When did you last build something with AI yourself that keeps running after you close your laptop?
3. Remove the Software Bill
Adam Hasemeyer, CEO of Valtech Americas, was paying seven figures a year in licensing for ERP, CRM, and finance tools, plus heavy labor costs to close the books.
His team started small. One minimum viable product, built in-house, to prove AI would do the job. Six months later they had custom software running for the executive team.
“Let’s have a goal of reducing licensing expenses… building more of these tools for ourselves to replace some of the manual, more laborious things that happen within the business.” – Adam Hasemeyer, CEO of Valtech Americas
They now get live visibility into performance and forecasting, shaped to how they run the company. They stopped paying for the parts they never used.
Pull your ten largest software contracts this week. For each one, ask an engineer what it would take to build the 20% of that product your people actually touch.
4. Remove the Sprawl
Every employee with an AI tool is now a software developer. That is the problem.
Jason Veiock, CEO of Bearing, warned me about a global workforce vibe coding private applications with no oversight, creating a web of disconnected tools that nobody owns. AI will only be as good as the context you give it.
“…what you’re going to see at the enterprise level is a consolidation around these systems of record because they’re going to provide the context… the governance, and on top of that governance comes the security.”
Jason Veiock, CEO of Bearing
Dun & Bradstreet surveyed 10,000 businesses in July. Only 6% said their data is fully ready to support AI at scale.
Consolidate AI use around the systems that already hold your data and your controls. Your CRM. Your ERP. Your document store. They know who your customers are and who is allowed to see what.
Then name the person who owns AI cost and AI sprawl in your company. More than half of the leaders in that July survey could not.
5. Remove AI From One Place
Here the pattern reverses.
Andrew Palosi, CEO of Ad Leverage, draws a hard line at customer-facing thought leadership. Google keeps shipping updates built to penalize repurposed AI content. Search authority rests on experience and trust. A company that swaps human originality for generated text will watch its visibility fall, and revenue with it.
“…to think that you can use AI to create… any type of original thought leadership and try to pawn that off as originality… is not only misguided, but it’s frankly, Glenn, it’s quite dangerous…” – Andrew Palosi, CEO of Ad Leverage
Use AI hard on the backend. Reconciliation, logistics, research, data, first drafts. Keep your name on the thinking.
What To Do This Month
Adding AI was the easy part. SAP and Oxford Economics surveyed 2,600 executives in July. 30% of all tasks now run with AI assistance, up from a quarter last year. Cost efficiency was not among the top benefits they reported.
Pick one thing you will remove. A software contract you will not renew. A role you planned to hire and will not. Give it an owner and a date, then report the result at your next leadership meeting.
Then ask the room what comes off the list after that.
