Only 37% of companies report any positive impact on earnings from their AI programs, according to McKinsey's Technology Trends Outlook 2026, even as 89% now deploy AI regularly and most are testing AI agents. The consulting firm's analysis reveals a substantial disconnect between widespread adoption and measurable economic returns, with the investments in agent-based systems outpacing their financial benefits. This gap becomes especially concerning given that AI agents demand considerably more resources and complexity than earlier AI tools.
A single agent workflow consumes five to 30 times the computing power of a standard chatbot interaction, the report calculates, because these systems rely on repeated inference loops, tool calls, retrieval operations, and coordination across multiple systems. This resource intensity has pushed 93% of survey participants beyond their planned AI budgets. In software development specifically—where agents hold the potential to unlock nearly a trillion dollars in value—only a quarter of companies using AI agents achieved meaningful acceleration in their development cycles. The study defines significant progress as at least doubling productivity for more than 25% of teams, a threshold most organizations haven't reached. Among individual developers, roughly 80% saw productivity gains of just 3% when using AI tools, while the top 20% experienced increases of 55%. Meanwhile, 30% of companies actually saw productivity decline after development teams began using coding agents.
The report finds that quality, not quantity, determines economic benefit from AI-generated code. In one analysis, programming activity surged 180% thanks to AI tools, but published releases grew only 30%, according to McKinsey. Trust remains another obstacle: approximately 46% of developers worldwide actively distrust the accuracy of AI tools, while just 3% express strong confidence in the results. The consulting firm writes that many teams are programming by "feel" rather than strategic intent, which doesn't consistently produce added value.
McKinsey acknowledges this poor track record doesn't have to persist permanently. Agentic AI remains in early stages, and the benefits companies extract from it will likely grow once the technology matures and integrates more deeply into enterprise operations. Recent development tools from Anthropic and OpenAI have enhanced tool usage and computer interaction, while frameworks like LangGraph from LangChain have introduced the orchestration and human-in-the-loop controls necessary to guide agents through extended workflows. Benchmarks from nonprofit Model Evaluation and Threat Research show growing capability: for Claude Opus 4.6, the organization estimates a 50% success rate on software engineering tasks that would take a human approximately twelve hours. Despite early struggles, capital investment is accelerating rapidly—funding for providers of agent-based software development solutions jumped from negligible levels in 2023 to around $5 billion in 2025, then exceeded $61 billion in the first half of 2026, driven primarily by SpaceX's $60 billion stock acquisition of Cursor. Job postings in this space increased 221% between 2024 and 2025, signaling sustained employer commitment even as returns lag investment. The analysts envision not merely faster individual coding but a complete redesign of the product development lifecycle, with agents analyzing requirements, generating code, conducting tests, identifying bugs, and managing portions of deployment. Organizations pursuing agents without clear business cases and measurable outcomes risk substantial costs with little payoff, as infrastructure upgrades become necessary and operational expenses mount from computationally intensive workflows that current budgets weren't built to support. The technology's promise remains enormous, but realizing it will require patience, strategic focus, and tolerance for a maturation period that separates experimental deployment from genuine competitive advantage.