Here’s what we learned about AI projects from enterprise buyers so far

Enterprise buyers are maturing quickly when it comes to AI and AI agents and looking to redesign operations, deliver outcomes, maintain governance and cost discipline. If there was a word to describe what enterprises want it's "optionality" since AI and the vendor landscape are changing to quickly to make a wrong bet.

Those are the lessons learned in the first half of 2026 where CxOs went from high on agentic AI buzz and OpenClaw banter to realism to outright sticker shock over token spend. It was hard to find CIOs who actually bought into tokenomics, but they were still left stunned with budget concerns as the first half ended.

Another thread for the first half was that CxOs are looking to architectures that will provide model options, deliver value and protect proprietary data with on-premise options. Whether vendors are speaking to these issues remains to be seen. The first half of 2026 was marked by a long list of private previews and chatter about outcomes from vendors and IT buyers decidedly wary of betting big on any one platform.

Constellation Insights went through 67 buy-side articles in the first half including Constellation Research's AI Forum and Futures Forum to mine the common themes. The shift in the first six months of 2026 is clear. With the new year, enterprises went from 'which AI model should we adopt?' to 'how do we redesign operations, customer experiences and technology architecture without surrendering control of our economics, data, or future choices?'

Part of that control resides in cybersecurity. Anthropic's Mythos moment sparked a run on cybersecurity efforts.

Here's a look at what we learned from the first half.

AI value has replaced the AI sizzle reel. Winning AI use cases are driven by processes and removing friction inside of the enterprise. Think supply chain, forecasting, scheduling, maintenance, inventory and all of those backend workflows.

Start with the operational pain points and deliver outcomes.

Customer facing AI still matters and garners its share of attention, but the returns so far have been about the workflows behind the scenes where AI can remove cost, time, errors and friction.

• For Starbucks, AI returns are behind the scenes
• How Moody's is thinking about data moats, AI strategy, token costs

ROI moves to the front with cost discipline close behind. Most companies are investing heavily in AI, but the burden of proof is rising. BT150 executives have said AI is at an operational inflection point even as they search for returns. CFOs are pressured to manage costs yet continue to spend on AI, data projects and cloud.

• BT150 zeitgeist: In search of AI agent ROI
• Where’s tokenomics for the rest of us?
• AI inference costs are going to be a big concern: What's the fix?

On the cost discipline side, AI is becoming a real budget line as AI agents proliferate. AI financial ops are on deck and keeping costs in line will lead to more on-premises deployments, model routing and, more importantly, AI business cases tied to outcomes such as revenue growth, margin improvement, cycle times and quality.

The AI foundation is driven by data, integration and context. The first half of 2026 was all about context from vendors to the point where the word is almost as meaningless as paradigm shift. Most AI projects include a data project riding shotgun.

Context will be about trusted proprietary and business-specific knowledge. Repeat after me: The model is not the differentiator. The grounding in your proprietary data is the differentiator. Here's the issue: It's increasingly difficult to sort out what's real and what's fantasy as every tech vendor introduces another context layer.

The best advice of the first half came from Virgin Australia, which has created a neutral architecture that revolves around data from its various systems and integration to power AI agents and automate processes.

• Two apartment rental giants merge, aim to leverage data, AI flywheel
• Caterpillar’s AI Vision Takes Shape, but Data, Cloud Focus Was the Precursor

Agentic AI is real, but deployment and scaling aren't easy. Verizon outlined how it manages agent sprawl. BT150 executives cite multiple growing pains. And any project that requires a small army of forward deployed engineers simply isn't ready for scale.

• AI Forum 2026: There isn’t an easy button for AI

What's missing? The maturity and homework needed ahead of time such as process architecture, workflow design, governance and that timeless need for change management.

Technology buyers want options and architecture control and lack trust in vendors. Enterprises are wary of becoming locked into any one model, application or platform. After all, enterprises have been locked in during multiple technology transitions from the mainframe to ERP to SaaS. We all know where this is headed: Your AI champion today is going to trap you for their own profit margin benefit tomorrow.

The term I heard repeatedly when talking about OpenAI and Anthropic was "Uber-ed." Everyone knows that enterprise AI is heavily subsidized to gain users and share. Uber did the same thing. We also know how that story played out as prices went higher as an ecosystem gains leverage. OpenAI and Anthropic are just newfangled plays on the Uber model. If you want another analogy you can substitute Uber for Netflix. Either way, we all know the drill: Things that look good today usually mean you’re screwed tomorrow.

Buyers aren't rejecting strategic vendor relationships, but they want optionality and more clarity. Right now, the AI winners and losers are unclear.

• Don't let your vendor's AI strategy become yours
• JPMorgan Chase CEO Dimon: We can't predict AI winners and losers yet

CEOs own AI success and failure. AI is a CEO and board-level agenda item and has graduated from a CIO or CTO initiative. Winning AI projects all have a common thread--CEO support and sponsorship.

• CEOs take control of AI projects: What could go wrong?
• Constellation Research's Futures Forum: What CEOs are thinking

Here's where this CEO as Chief AI Officer theme gets muddled. AI agents require a new operating model. AI can't be separated from organizational design and CEOs need a strong team. My bet: There will be big CEO wins from AI deployments. There will also be spectacular AI failures.

• The CEO’s AI Dashboard (a series begins)
• CEO's AI Dashboard Series: Business Value (#2)
• The CEO’s AI Dashboard, Part 3: Data Readiness
• The CEO’s AI Dashboard, Part 4: Infrastructure Readiness

AI may require more humans. The year started with the storyline that AI agents will replace humans. In fact, tech bros couldn't stop talking about how AI agents are going to replace human labor. They've backpedaled quickly now they see the political backlash to AI.

We'll set the Silicon Valley types aside and roll with the people in the field. The consensus from the BT150 is that AI is going to require more people. Someone has to manage all of these AI agents that'll run amok if you let them.

• Your AI talent planning algorithm is common sense

The lesson here may come from Starbucks, which is using AI in the background to make its employees more effective. Starbucks is betting on the human connection.

We have already seen enterprises start to rev up the college hiring efforts. There will be workforce upheaval due to AI, but we're not talking wholesale labor substitution.

The bottom line

The buyer market is becoming more sober and more sophisticated. Enterprises are still investing, but the strongest programs share a common pattern:

• Start with a business workflow and what needs to be done regardless of the model.
• Build on trusted data, integration and reusable context.
• Measure outcomes and token economics from the beginning.
• Govern agents as production systems, not experiments.
• Preserve architectural control and model and vendor flexibility.
• Use AI to augment people and improve customer experiences, not merely automate tasks. We've over-rotated on the idea AI agents will replace people.
• Treat AI as a multiyear operating-model transformation led by the C-Suite and business unit leaders. The CEO is the ultimate owner.

Themes for the second half of 2026

Making any predictions in enterprise AI is pure lunacy, but running headfirst into a wall is a bit cathartic. Here’s what I’m expecting for the rest of 2026.

• Governance will move to the front. Why? There will be some massive agentic AI screwup that'll put some urgency into governance.
• One large vendor will actually pull back on AI infrastructure spending and that'll create a cascading effect.
• The industry will move beyond the AI factory and tokenomics narrative as everyone realizes we're basically creating a mainframe model for AI.
• Model routing will become critical as enterprises focus on AI's cost. FinOps and AIOps are going to converge.
• The IPOs of OpenAI and Anthropic will be the peak of the AI insanity and usher in the realism era (at least until quantum computing, space or humanoid robotics take over the hype baton).
• Once we're all sick of AI and almost disgusted by it the real returns on investment will come.
• We’ll all agree that SaaS isn’t dead, but we still won’t think highly about our vendors and the stocks will tread water at best.