McKinsey just published a report on one year of agentic AI.
Their takeaway: value comes from redesigning workflows, and agents aren’t always the answer. Many use cases are better served by rules-based automation, analytics, or GenAI; deploy agents when the work involves multistep decisions with lots of variance and judgment.
And to be fair, this isn’t on executives themselves. Much of what the market is selling as “agents” are really just rules-based automations stitched together with GenAI. They look slick in a demo, but they’re closer to if-this-then-that systems with a chatbot on top.
I’ve written and talked about this before, and I believe this mismatch is what will hurt AI adoption.
Businesses expect transformation, but when the P&L doesn’t move, budgets get cut and momentum stalls. Not because AI failed, but because what they bought wasn’t agency at all.
Automation in Disguise
Workflows (rules-based automation): If-this-then-that logic, stitched together with AI helpers. Reliable, predictable, but rigid.
Agents: Given a goal, they figure out the “how”. They reason, adapt, use tools, and change course when conditions shift.
That’s the difference between building a train (fast but stuck on rails) and a car (chooses its own route). Right now, most businesses are laying more train tracks and pretending they’ve built cars.
Why Agents Could Be Game-Changing
McKinsey’s latest analysis makes the distinction clear. In a complex workflow (like an investigation), different tools shine at different points:
Screenshot from McKinsey Report.
The point is that agents aren't a replacement for rules-based automations, analytics, or GenAI, they sit above them. Workflows execute instructions; agents pursue goals.
An Example of Where Each Fits
Rules-based automations: Best for repeatable, standardised tasks where steps never change.
Insurance claim example:
Route the claim to the right department.
Reads a policy document for anomalies
Extract records from a database.
Notify the claimant once steps are complete.
Outcome: the claim is moved along the process efficiently. It saves manual admin time, but the system doesn’t know if the claim is valid, urgent, fraudulent, or fair, it just follows the rails.
Agents: Best when the outcome isn’t predetermined and adaptation is required.
Example: In the same insurance claim, an agent could:
Review the claim narrative and supporting documents.
Draft a work plan for additional evidence needed.
Through its own determination, cross-checks past cases, spots anomalies, and generates clarifying questions that it converses with the claimant back-and-forth without human intervention.
Synthesises findings and recommends whether the claim should be approved, denied, or escalated for a human to sense check.
Outcome: the claim isn’t just moved forward, it’s assessed. The agent reduces cycle time, spots risks (like fraud or missing coverage), and produces a reasoned recommendation with supporting evidence. Instead of waiting for multiple handoffs, the insurer can resolve straightforward claims quickly and focus human experts on the truly complex ones.
The difference is huge: the workflow moves the file along; the agent moves the decision forward.
McKinsey’s point is that complex workflows should combine the best tools at each stage. Rules handle routing, analytics flag risks, GenAI drafts documents, and agents orchestrate towards outcomes. Misapplied, you get disappointment: deploying “agents” where rules suffice, or worse, calling workflows “agents” and expecting transformation.
The Agent Illusion and the Budget Backlash
If businesses mistake workflow automation for agency, they’ll think they’ve transformed when all they’ve really done is patched the old system. And when the bottom line doesn’t move, the backlash begins.
Executives have been sold the story that “agents” will unlock transformational productivity, collapse costs, and create new capacity. They expect to see a step-change in the P&L. But when the dust settles, what’s been rolled out are rules-based automations dressed up with AI, branded as agents, and priced like transformation. The bottom line doesn’t move.
Why This Happens
Automation disguised as agency: shaving minutes off tasks instead of changing workflows.
Tech as panacea: throwing tools at problems without redesigning processes.
Agents sold as strategy: when in reality they’re just workflow optimisers.
The Cost of the Trap
When the promised shift in the P&L doesn’t arrive, leaders don’t ask “Did we redesign the right workflow?” They ask “Why are we spending so much on AI?” That’s when budgets collapse, momentum stalls, and credibility is lost.
We’ve seen this cycle before. After the iPhone launched, every brand rushed out to build an app. Most were clunky web wrappers duplicating websites. They didn’t drive revenue or loyalty. The winners weren’t the apps themselves but the new business models they enabled: Uber, Airbnb, Instagram. That was the real “there’s an app for that” moment.
This Will Hurt the AI Industry
If leaders keep confusing automation for agency, two forces work against them and the industry:
Wasted investment: forcing AI into problems a simple rules engine could solve.
A hard ceiling: optimising the old model instead of reimagining what’s possible.
Point 2 will be a significantly limiting factor for many companies. They’ll keep optimising workflows built for the legacy business era, defaulting to automation over agency, shaving minutes rather than reimagining outcomes that shift the P&L or even the entire business model.
And that’s how market share gets eaten. It won’t be the nearest competitor that takes them down.
Just like how...
Blockbuster wasn’t killed by Hollywood Video but by Netflix rewriting the model, or
Barnes & Noble lost ground not to Borders but to Amazon rethinking distribution, or
The accounting incumbents who were blindsided when Xero put enterprise-grade capability into the hands of every small business.
Today’s incumbents risk the same fate. The threat comes from the Stanford spin-out or the two kids in a dorm who aren’t burdened by legacy thinking. They don’t optimise the old world, they design the new one.
That’s the rude awakening coming fast for leaders who slap “agent” on rules-based automation.
What Leaders Should Do
Take McKinsey’s advice: don’t buy the hype, redesign the workflow. The companies that see real value will:
Start with workflows, not agents. Map the process end-to-end and ask where speed, accuracy, or decision-making truly break down.
Match the tool to the task. Use rules for predictable steps, analytics for scoring, GenAI for summarising or drafting, and agents only where outcomes require reasoning and adaptation.
Treat agents like hires. Define their role, test their performance, and give them guardrails and feedback loops.
Make it observable. Don’t just track final outcomes, measure how each step is performed so errors surface early.
Reuse and scale. Avoid bespoke one-off builds; share prompts, agents, and workflows across teams.
Keep humans in the loop. People validate, sense-check, and oversee. Trust comes from transparency, not blind delegation.
Leaders who follow these steps will capture the gains McKinsey describes: cycle time cut, costs collapsed, new capacity unlocked. Those who don’t will keep laying more train tracks while the upstarts in dorm rooms and Stanford labs design the cars that overtake them.
The Next Step
The next step then is breaking through the artificial ceiling, reimagining not just workflows but the very business model through the lens of AI. That’s the leap McKinsey points to: don’t stop at faster execution of the old model, ask what new model AI makes possible. To avoid that, leaders need to reimagine the business itself and ask:
If we rebuilt from scratch today with AI at the core, what would we design differently?
Which processes would never exist if agents could deliver outcomes end-to-end?
What new products, services, or revenue streams become possible with AI that weren’t before?
Which parts of our model will competitors (or upstarts) rewrite first if we don’t?
Leaders willing to ask and act on those questions will avoid the ceiling and capture the real prize: AI not as a faster train, but as the chance to redesign the entire journey.
Written by Mike ✌
Passionate about all things AI, emerging tech and start-ups, Mike is the Founder of The AI Corner.
Subscribe to The AI Corner
