The State of AI in 2026: Closing the Gap Between Adoption and Impact

What the Research Says

McKinsey's annual State of AI survey (August 2026 edition, "On the Road to ROI") gives a useful reality check on where enterprise AI actually stands.

Adoption Is Near-Universal, But Profit Impact Hasn't Followed

• About 90% of organizations now use AI somewhere in the business.
• 44% are scaling it enterprise-wide  up sharply from prior years.
• Agentic tools, especially coding agents, are moving from novelty to default in many engineering teams.
• Yet only 37% of companies attribute any profit impact to AI, a number that's barely moved in a year.
• Only a small slice of "high performers" are seeing anything close to transformational returns.

The Human Side Tells a Different Story

• 8 in 10 employees say AI has genuinely made them more productive a felt, day-to-day benefit, not hype.
• The gap: that personal productivity gain hasn't yet translated into something a CFO can point to on a balance sheet.

What Separates the High Performers

• ~75% of high performers have redesigned workflows around AI.
• Only ~25% of everyone else has done the same.
• McKinsey's read: this not tool count is the deciding factor.

On Workforce Impact

• Only 14% of organizations saw actual headcount declines this year well under half of what was predicted twelve months ago.

Zymr's POV

We work inside client organizations that are, for the most part, exactly where this report describes: past the pilot stage, not yet at enterprise-level returns. As a software engineering services company, we'll keep our take focused on where we actually build things workflow and system redesign, and the build-vs-buy shift in engineering not on broader workforce or economic commentary.

On the Redesign Gap

This is the single most important number in the report, and it matches what we see on the ground: companies bolting an AI feature onto an existing process rarely get past the pilot. The ones pulling ahead have rebuilt the actual unit of work the workflow, the handoffs, the approval steps around what the AI can now do differently, not just faster.

• Dropping a copilot into an existing process ≈ marginal productivity gain, hard to attribute to profit.
• Redesigning the process around agentic capability ≈ where the returns McKinsey calls "transformational" actually show up.
• This is systems and process engineering work first, model selection second and it's usually where client engagements stall if it's skipped.

This is systems and process engineering work first, model selection second  and it's usually where client engagements stall if it's skipped.

On the Build-Versus-Buy Shift in Engineering

We're seeing this directly with clients: as coding agents improve, more teams are choosing to build functionality in-house that they'd have licensed or outsourced a year ago. This is a real, fast shift in how engineering budgets get allocated faster than most procurement processes are built to handle.

• Our consistent recommendation: treat agentic coding as a capability requiring governance, cost visibility, and clear ownership not a tool you flip on and walk away from.
• Without that, "build in-house" quietly turns into unmanaged technical debt and untracked AI spend across teams.

On the Productivity-to-ROI Gap Specifically

The 8-in-10 productivity number is real, but individual productivity gains don't automatically aggregate into enterprise ROI unless the surrounding workflow, data pipeline, and system architecture are redesigned to capture them. That's the engineering gap between "people feel faster" and "the P&L shows it" and it's the gap we get called in to close.

Bottom Line

Scaling AI tools is the easy part. The hard part  the part that shows up in the numbers is having the engineering discipline to redesign the workflow and the systems around them, and to manage agentic coding as a governed capability rather than a novelty. That's the work that separates this year's high performers from everyone else still waiting for the ROI to show up.