85% Want Agentic AI. 76% Can't Run It Yet. The Problem Is Your Org Chart, Not Your Model.

Via MIT Technology Review: Rethinking organizational design in the age of agentic AI

Ambition outran the operating model

Enterprise AI has a familiar failure mode: buy the tool, assign a pilot, declare transformation, then wonder why nothing moved. A May 2026 sponsored report from MIT Technology Review Insights — produced with agent platform vendor Ema and featuring commentary from PwC's UK consulting leadership — puts numbers on that gap. In survey data cited in the piece, 85% of organizations say they want to be "agentic" within three years, while 76% say their current operations and infrastructure cannot support that change.

That is not a model problem. It is an operating-model problem — and it scales down to owner-led firms faster than most consultants admit.

The report's central metaphor is blunt: many companies are taping autonomous agents onto org charts built for humans — approval chains, handoffs, activity metrics, and siloed apps that made sense when work moved at human speed. PwC's Prasun Shah, quoted in the piece, compares it to adding sticky tape to a structure that is already cracking. Agents can execute whole workflows, coordinate across systems, and adapt without a manager in the loop. Bolting that capability onto a hierarchy designed for industrial-era throughput is how you get expensive pilots that never graduate.

Disclosure: The underlying MIT Technology Review Insights report was produced as sponsored content in partnership with Ema. This article independently summarizes and interprets it for SME owners; it is not affiliated with MIT editorial.

Why "add agents" is the wrong starting point

The sponsored report introduces a framework Ema calls agentic business transformation (ABT) — less a product pitch than a vocabulary for a shift the piece argues existing terms miss. Digital transformation moved paper to software. Copilots assist humans inside existing processes. ABT, as described, means weaving agents into how work is designed — not as another SaaS tile on the dashboard.

Three pillars recur throughout the analysis, and each one translates to SMEs without the enterprise jargon:

• Technology architecture. Legacy stacks assume humans click through linear workflows. Agents operate at machine speed across multiple systems at once. The report argues value comes when agents act as connective tissue — pulling context from CRM, billing, inventory, and email to make a decision — not when they sit inside one app waiting for prompts.
• Workforce design. Standard hierarchies optimize for standardized tasks and clear escalation paths. Agents that execute and coordinate without managerial babysitting blur those lines. Managers do not disappear; their job shifts from chasing status updates to governing trust, explainability, and hybrid-team dynamics. McKinsey figures cited in the piece — that roughly three-quarters of current jobs may need redesign, upskilling, or redeployment by 2030 — are enterprise-scale, but the direction applies anywhere headcount is thin.
• Success metrics. Activity metrics (calls handled, tickets closed, reports filed) break when an agent clears a thousand interactions in the time a human clears ten. The report pushes outcome metrics instead: retention, revenue impact, error rates, escalation quality. One enterprise example in the piece claims ROI measured on outcomes tripled within two quarters after ditching vanity throughput stats — take that as directional, not a guarantee, but the logic holds for a 40-person firm too.

What a 50-person firm should actually redesign

You do not need a reorg deck from a Big Four firm to act on this. You need honest answers about where agents are taped on versus where work was rebuilt around them.

• Pick one workflow, not ten tools. Customer onboarding, invoice exception handling, compliance doc prep — choose a process with clear inputs, measurable outcomes, and painful handoffs. Agents earn their keep on loops, not demos.
• Redraw decision rights before deployment. Who approves spend? Who owns a wrong answer? Who can pause an agent? The MIT piece raises exactly these questions for enterprise boards; your version fits on one page and should exist before go-live.
• Rewire data access, not just API keys. An agent that cannot see the same systems a good employee would check will hallucinate confidence. Architecture for SMEs is often "give it read access to these five sources, write access to none until reviewed."
• Replace activity KPIs with outcome KPIs. If your team celebrates "AI drafts completed" instead of "disputes resolved without rework," you will optimize the wrong thing. Outcome metrics also make ROI conversations with ownership possible.
• Plan for hybrid management. Your office manager, ops lead, or partner will suddenly supervise humans and agents. That requires training on escalation, logging, and when to override — not just a lunch-and-learn on prompting.

The 85%/76% split is the headline. The actionable lesson is smaller: agentic AI is an org-design project that happens to use software, not a software project that happens to touch org charts.

Someone has to own the operating tempo

Large firms hire chief AI officers and consulting armies. Owner-led SMEs hire none of that — and then wonder why agents sprawl across personal ChatGPT accounts while the "official" pilot gathers dust.

This is the work AgentsROI.ai does.

• A Workflow ROI Audit finds where agentic automation actually saves time and money — and where you are taping agents onto processes that need redesign first.
• Managed AI Operations keeps governed agent workflows running after the workshop ends: monitoring, fallbacks, spend controls, and updates so the operating model does not decay when the enthusiast goes on vacation.
• A Fractional AI Officer gives you senior judgment on vendor choices, decision rights, and outcome metrics without a six-figure hire — the role the MIT piece assumes exists in every enterprise, but rarely does in a 30-person shop.

AgentsROI.ai does not sell org-restructure theater. It sells practical operating design — vendor-neutral, sized for firms that cannot afford to learn this lesson twice.

Stop taping agents onto a human org chart

The MIT-sponsored analysis is enterprise-flavored, but the diagnosis travels: most organizations want agentic capability faster than they can support it, because they never redesigned how work flows, who decides, or what success looks like.

If you are an SME owner hearing "we should deploy agents" from your team, your first question is not which vendor. It is which process, which outcomes, and who owns the result when the agent gets it wrong.

Book a Workflow ROI Audit and find out whether you need more AI — or a simpler operating model that AI can actually run.

This article summarizes and interprets a May 26, 2026 sponsored report from MIT Technology Review Insights, produced in partnership with Ema and featuring commentary from PwC UK Consulting. The underlying report was sponsored content and was not written by MIT Technology Review's editorial staff. Statistics, frameworks, and quotes cited here reflect that source at the time of writing and may change. This article is for general informational purposes only and does not constitute legal, tax, financial, investment, security, or compliance advice. AgentsROI.ai is not a law firm, accounting firm, or registered investment adviser. Readers should verify current information independently and consult qualified professionals regarding obligations specific to their industry, jurisdiction, and circumstances — including applicable New York State and New York City requirements. AgentsROI.ai may have commercial relationships with vendors mentioned; where material, such relationships are disclosed. Nothing in this article is an endorsement of any specific AI product, model, or provider.