Your AI spend jumped 47% this year. Your CFO wants to know what that money bought. Your CTO can't answer. That's the $2.59 trillion enterprise AI problem Harness just built two products to solve.

The DevOps platform company announced AI DLC Insights and cloud & AI Cost Management on May 28, 2026. Together, they give engineering organizations real-time visibility into every dollar spent on AI—from the tokens developers burn writing code to the inference costs production agents rack up serving customers.

According to Gartner, worldwide AI software spending will hit $2.59 trillion in 2026, a 47% increase over 2025. Yet Harness's own State of Engineering Excellence report found that 94% of engineering leaders say the metrics that matter most are missing from their current measurement frameworks.

Trevor Stuart, SVP and GM at Harness, puts it bluntly: "The first phase of AI adoption was about getting teams to use and understand the tools. The next phase is about proving the tools have a positive impact. Demonstrating ROI will be the defining challenge of enterprise AI in 2026."

The CFO Problem: $2.59 Trillion in Spending, Zero Visibility

Here's what CFOs see today: A line item that keeps growing. Monthly invoices from OpenAI, Anthropic, GitHub, Cursor, and a dozen other AI vendors. Usage is up. Costs are up. But business outcomes? Unclear.

Traditional capital investment models don't work for AI. You're not buying servers you can depreciate over five years. You're buying tokens consumed in milliseconds. You're paying per-request for agents that might resolve customer tickets or might hallucinate nonsense.

Finance teams are asking questions engineering can't answer:

• What percentage of our AI-generated code actually ships to production?
• How much are we spending on abandoned code or bloated prompts?
• What's the cost per resolved support ticket when agents handle tier-1 issues?
• Which teams are burning tokens on experiments versus shipping features?
• Are we wasting money on expensive models when cheaper ones would work?

Most enterprises track AI spend at the invoice level. That tells you which vendor charged you $427,000 last month. It doesn't tell you if that money drove $2 million in efficiency gains or funded a hackathon project that shipped zero production code.

Harness's research reveals the scale of this gap: 94% of engineering leaders admit they can't measure what matters. They know developers are using AI coding assistants. They know production agents are handling customer workflows. But connecting spend to outcomes? That capability doesn't exist in most organizations.

This creates a governance crisis. Without measurement, you can't optimize. Finance can't build business cases for scaling AI investment. Engineering can't identify which tools deliver value and which ones waste money. And both sides end up making decisions based on anecdotal evidence instead of data.

The CTO Problem: Pilots Succeed, Production Costs Explode

CTOs face a different but equally painful challenge: Scaling AI from pilot to production without blowing up the budget.

A pilot project with 10 developers costs $5,000 per month. You roll it out to 500 engineers, and the bill hits $180,000. Did productivity increase proportionally? Are developers shipping 36x more code? Almost certainly not. But without instrumentation, you're flying blind.

Here's what makes AI cost management harder than traditional cloud FinOps:

Usage-based pricing at scale. Cloud infrastructure costs scale somewhat predictably with load. AI costs scale with every interaction—every line of code generated, every prompt submitted, every customer question answered by an agent. One chatbot handling 100,000 conversations per month can rack up inference costs faster than the revenue it protects.

Productivity leakage. Developers save 30 minutes per day using AI coding assistants. Great! Except they're not reallocating that time to strategic work. They're using it to write more code—often experimental code that never ships. You're paying for tokens that produce zero business value.

Wasted spend on abandoned code. A developer uses Claude Code to scaffold a feature, burns 50,000 tokens, then pivots to a different approach. That original code never makes it into a pull request. Traditional development tools don't track this. AI cost management systems need to.

Model selection chaos. Your most expensive AI models (GPT-4, Claude Opus, Gemini 1.5 Ultra) cost 10-50x more than cheaper alternatives. Developers default to the best model for every task, even when a faster, cheaper one would work fine. Without visibility into which models teams are using—and whether those choices are justified—you're leaving money on the table.

No connection between development costs and production outcomes. Even when you track token spend during development, you can't tie it to production impact. Did that AI-generated code ship faster? Did it reduce incidents? Did it improve DORA metrics? Most organizations have no idea.

Harness built AI DLC Insights and Cloud & AI Cost Management because its own customers were hitting these problems at scale—and existing tools couldn't solve them.

AI DLC Insights: Tracking Code from Prompt to Production

The development-side product is AI DLC Insights, which extends Harness Software Engineering Insights with a new on-machine developer agent.

That agent runs directly in the developer's environment and captures every AI-generated line of code. It records token costs per model and per tool (Claude Code, Cursor, GitHub Copilot, Windsurf). And it maps that spend through the full delivery chain—to the pull request, ticket, and deployment it produced.

The result is a complete picture of developer AI ROI:

Unified adoption visibility. A single dashboard shows which AI coding tools your teams actually use, how often, and at what cost. No more hunting through vendor invoices to figure out who's using what.

Per-developer attribution. Token spend, active sessions, and shipped code are traced to the developer, team, and business unit. You can benchmark high-performing teams against org-wide averages and identify outliers.

Wasted spend detection. The system surfaces abandoned code (tokens spent on work that never makes it into a PR), bloated prompts (inefficient requests that burn tokens unnecessarily), expensive model choices (using GPT-4 when GPT-3.5 would work), and missed cache hits (re-generating responses instead of reusing cached results).

Coding-to-production impact. AI DLC Insights tracks AI-generated code from prompt to production using ship rate, pull request cycle time, and DORA metrics (deployment frequency, lead time for changes, change failure rate, mean time to recovery). It correlates those metrics with incident data to answer the critical question: Is AI-assisted code better or worse than human-written code?

Benchmarking and governance. Teams can compare their performance against org-wide baselines. Role-based access control ensures managers see aggregate data while protecting individual developer privacy.

This is the instrumentation CTOs need to justify continued investment in AI coding tools. If you can prove that developers using Claude Code ship 20% faster with 15% fewer incidents, the ROI case writes itself. If you discover teams are burning $40,000 per month on abandoned code, you've identified a $480,000 annual savings opportunity.

Cloud & AI Cost Management: Unit Economics for Production Agents

Once AI ships to production, the cost equation changes. Now you're paying for inference—every customer interaction, resolved ticket, automated workflow, or agent-driven decision.

Most organizations only see this spend at the invoice level. You know AWS Bedrock charged you $83,000 last month. You don't know if that was 10 million low-value queries or 500,000 high-value automation workflows.

Cloud & AI Cost Management makes that distinction possible. The product extends Harness's existing Cloud Cost Management platform to cover every dollar of AI infrastructure spend, connecting directly to AI providers (OpenAI, Anthropic, AWS Bedrock, GCP Vertex AI) and production agents to capture spend at the individual request level.

Key capabilities include:

Unified AI cost visibility. A single view of spend across every AI provider and managed service provider, eliminating the need to consolidate invoices manually.

Full spend attribution. Costs are traced to the agent, model, team, and business unit driving them. You can calculate unit economics: cost per resolved support ticket, cost per completed sales workflow, cost per customer onboarding session.

Anomaly detection. Unusual AI spend spikes are flagged proactively—before they hit your invoice. If a production agent suddenly starts burning 10x its normal inference budget, you know within hours, not weeks.

Budget and governance controls. You can set spending limits at the agent, team, or business unit level, extending existing FinOps controls to AI spend. If a team hits 80% of their monthly AI budget, the system can alert them or enforce throttling.

This is what CFOs need to build business cases. If you can prove that an AI-powered support agent resolves tickets at $2.50 per resolution versus $45 for human agents, scaling that agent from 10,000 tickets per month to 100,000 becomes a no-brainer.

Why This Matters: From R&D to ROI

The enterprise AI narrative is shifting from "Can we do this?" to "Should we keep paying for this?"

In 2024 and 2025, enterprises ran pilots to prove AI was feasible. CTOs got budget to experiment. Engineering teams explored new tools. Finance approved spending because "everyone's doing AI."

2026 is different. Boards want results. CFOs want payback periods. CEOs want proof that AI investments are driving revenue, reducing costs, or creating competitive moats.

Harness's products address this shift directly. AI DLC Insights gives CTOs the data to prove coding assistants improve velocity and quality. Cloud & AI Cost Management gives CFOs the unit economics to justify scaling production agents.

Without these tools—or something equivalent—enterprises are making multi-million-dollar AI bets with zero instrumentation.

Consider two scenarios:

Scenario A (No Measurement): Your organization spends $5 million per year on AI tools and infrastructure. Developers report "it's helpful." Support tickets are "probably getting resolved faster." Leadership approves another $5 million for 2027 based on vibes and anecdotal evidence.

Scenario B (Full Measurement): You spend $5 million, but you know that $1.2 million went to abandoned code, $800,000 to inefficient model choices, and $500,000 to experimental projects that never shipped. You also know that the remaining $2.5 million drove $12 million in efficiency gains (faster shipping cycles, fewer incidents, automated tier-1 support). You cut the waste, double down on what works, and hit $20 million in ROI next year.

That's the difference between hoping AI delivers value and knowing it does.

What This Means for Enterprise Leaders

For CFOs:

• You can finally answer the board's question: "What did we get for our AI spending?"
• You can build data-driven business cases for scaling AI investment
• You can identify waste and reallocate budget to high-ROI initiatives
• You can set spending limits that prevent runaway costs without killing innovation

For CTOs:

• You can prove which AI coding tools deliver measurable productivity gains
• You can identify teams that ship AI-generated code 30% faster with fewer bugs
• You can cut spending on abandoned code and inefficient model choices
• You can track production agents from cost per request to cost per business outcome

For VPs of Engineering:

• You can benchmark team performance and share best practices across the organization
• You can justify headcount requests by proving AI multiplies team output
• You can eliminate tools that look productive but don't drive shipping velocity
• You can correlate AI adoption with DORA metrics to measure code quality impact

The Bottom Line

Harness's AI DLC Insights and Cloud & AI Cost Management are available in beta today. The company is positioning them as the instrumentation layer enterprises need to move AI from experimental spend to measurable ROI.

Will these products become the standard for AI cost management? That depends on how quickly Harness can onboard customers and prove their own ROI claims. But the problem they're solving is real, urgent, and growing at 47% per year.

If your organization is spending six or seven figures annually on AI—and you can't answer basic ROI questions—these tools deserve a serious look. The alternative is continuing to fly blind while your AI budget compounds.

And in 2026, "everyone's doing AI" is no longer a sufficient answer when the CFO asks what you're getting for all that money.