AI Transformation: Overlay Agents or Redesign the Org?
Enterprise leaders are making a specific AI transformation decision in 2026: should they add AI agents to existing workflows for faster near-term gains, or redesign operating models so agents can own meaningful parts of work? The distinction matters because the market is showing a wide gap between ambition and readiness. According to MIT Technology Review Insights, 85% of organisations want to become agentic within three years, yet 76% say their current operations and infrastructure are not ready.
That gap suggests many enterprises are not facing a tooling problem first. They are facing a design problem: how technology, management, and measurement need to change when AI stops acting like an assistant and starts acting more like an operator across workflows.
Overlay vs. redesign: the real choice in AI transformation
| Criterion | Add agents to current workflows | Redesign the operating model for agents |
|---|---|---|
| Time to first pilot | Faster, often measured in weeks | Slower upfront because process ownership must be clarified |
| Scope of value | Narrow productivity gains in one team or workflow | Broader gains across functions and handoffs |
| Architecture needs | Can work on top of existing apps with limited integrations | Requires stronger enterprise AI integrations across systems and data |
| Management impact | Minimal org change at first | Managers and process owners need new roles and controls |
| KPI model | Usually output metrics such as tickets handled or reports generated | Outcome metrics such as cycle time, escalation rate, conversion, or retention |
| Failure mode | Point solutions, duplicated steps, unclear accountability | Slower rollout, but cleaner scale if governance and ownership are set |
The market is increasingly splitting along these two models. The overlay path is attractive because it fits annual planning cycles, existing budgets, and familiar approval structures. But it also tends to preserve the same handoffs, same hierarchies, and same reporting lines that limited earlier digital transformation AI programs.
The redesign path demands more from leadership. It requires decisions on workflow ownership, cross-functional data access, and where humans retain approval rights. That makes it harder to start, but it is also the path more aligned with end-to-end AI business automation rather than isolated experiments.
Why the sticky-tape model breaks down
MIT Technology Review’s reporting centres on a point made by PwC UK Consulting’s Prasun Shah: many firms are still embedding AI employees into what is essentially a human operating model. He compared that approach to adding sticky tape to parts of an operating model that is already breaking.
That trade-off is straightforward. Layering agents onto old processes can produce visible wins in customer service, HR, or sales, especially where work is repetitive. The source notes estimates that AI agents could accelerate business processes by 30% to 50% and reduce low-value work time by 25% to 40% at scale. Those are meaningful numbers. But they can also mask structural friction if the surrounding workflow remains linear, approval-heavy, and fragmented across applications.
A comparative reading of the market shows three common reasons the overlay model stalls:
- Agents inherit bad process design. If the underlying workflow has redundant checks or unclear ownership, the agent just executes confusion faster.
- Enterprise AI integrations stay shallow. An agent limited to one system cannot coordinate the broader job.
- Teams measure activity, not value. High task volume can look impressive while business outcomes barely move.
This is where AI strategy starts to matter more than model selection. The useful question is not only which agent platform to buy, but which workflows are worth rewiring so that agents can coordinate work across systems instead of adding another interface layer.
Technology stack vs. connective tissue
Ema’s framing, covered in the source article, is useful because it treats agents not as another application but as connective tissue moving across systems. That is a different architectural assumption from the application-centric stacks most enterprises built in the last decade.
In the overlay model, AI workflow automation usually sits inside a narrow task boundary: summarize a case, draft a response, classify a form, route an exception. That can be productive, and in some environments it is the right first move. The trade-off is that each automation remains dependent on human coordination between systems.
In the redesign model, agents are configured to retrieve context from multiple systems, interpret it, and complete a larger business task. That is closer to the source article’s description of agents executing entire workflows with limited human input. It is also why architecture becomes decisive. As McKinsey’s work on generative AI and the next productivity frontier has argued, value rises when AI is embedded in core processes rather than parked at the edge.
The trade-off here is speed versus durability. Overlay automation can start with lighter integration work. Redesign needs stronger data access, better process maps, and more deliberate AI implementation services. But if an enterprise wants agents to move from pilot to production without six months of custom software work for every use case, connective-tissue architecture is the better long-term bet.
A relevant internal reference point is Encorp’s service page for AI Integration Services for Microsoft Teams. It is not a full operating-model redesign offer, but it fits the training-stage discussion because it shows how workflow-level AI integration can expose where collaboration patterns and process ownership need to change before broader rollout.
Hierarchies vs. hybrid teams
The workforce comparison is just as important as the technology comparison. Traditional org charts assume that coordination, escalation, and optimisation move through layers of human managers. Agentic systems weaken that assumption.
According to the source, Shah argues that managers in hybrid teams will need to handle trust, explainability, psychological safety, and status dynamics. That suggests a shift in management work from supervising execution to supervising judgment, exceptions, and accountability.
| Workforce question | Legacy hierarchy | Hybrid human-agent team |
|---|---|---|
| Who executes routine work? | Analysts, coordinators, agents in the HR sense | Software agents plus human reviewers |
| What do managers do? | Assign tasks, monitor output, escalate issues | Set guardrails, review exceptions, resolve conflicts, monitor outcomes |
| How is capability built? | Hiring and training by function | Upskilling, redeployment, and workflow redesign across functions |
The trade-off is not humans versus machines. It is whether an enterprise is ready to redesign jobs around orchestration, exception handling, and decision quality. McKinsey has estimated that by 2030, a large share of current jobs will require redesign, upskilling, or redeployment. In practical terms, that means custom AI agents are not simply a procurement decision; they are a staffing and operating-model decision.
Output metrics vs. outcome metrics
This may be the most underappreciated comparison in current AI transformation programs. Output metrics flatter early deployments. Outcome metrics expose whether the system is actually improving the business.
Ema’s example in the source article is telling: one enterprise shifted from tool metrics such as cost per query and model accuracy to business outcomes such as the percentage of contracts reviewed without human escalation, and reported that measured ROI tripled within two quarters. Whether that exact gain generalises is less important than the principle. If the KPI system stays tied to activity, AI will optimise the wrong target.
When you add AI employees into the workforce, activity metrics become meaningless or actively misleading, Ema CEO Surojit Chatterjee told MIT Technology Review Insights.
The comparison is clear:
- Output metrics help when the goal is testing technical reliability.
- Outcome metrics help when the goal is operational and financial performance.
A useful benchmark comes from Gartner’s guidance on driving positive ROI on AI, which emphasises linking AI initiatives to business outcomes rather than isolated technical indicators. For enterprise buyers, this is where many AI implementation services engagements either create discipline or create reporting theatre.
What leaders should redesign first
The evidence from the source article, and from broader enterprise AI adoption patterns, points to a sequencing question rather than a binary yes-or-no decision. Not every workflow needs a full redesign on day one. But enterprises do need to know which layer they are changing first.
A workable sequence looks like this:
- Pick one cross-functional workflow, not one tool. Customer onboarding, contract review, HR case handling, and sales operations are stronger starting points than single prompts or assistant features.
- Map the handoffs before buying more agents. If ownership, escalation paths, and required systems are unclear, the pilot will produce noise.
- Set outcome KPIs before rollout. Cycle time, escalation rate, first-pass completion, and revenue or retention effects matter more than activity counts.
- Train managers for hybrid supervision. This is why the program-stage fit here is leadership education first, then deeper implementation.
The broader implication is that AI transformation is becoming less about adding intelligence to tasks and more about redesigning how work is coordinated. That is a more demanding agenda than most 2024-era copilot projects, but it is also where durable value is likely to accrue.
Verdict: pick the overlay model if the goal is a fast pilot, a narrow workflow, and low organisational disruption. Pick the redesign model if the goal is enterprise-scale AI workflow automation, stronger enterprise AI integrations, and a KPI system that measures outcomes rather than activity.
Martin Kuvandzhiev
CEO and Founder of Encorp.io with expertise in AI and business transformation