Designed Friction by Artificial Intelligence
Why AI Productivity Needs More Control, Not Less.
AI is being sold on speed.
Faster search. Faster service. Faster content. Faster shopping. Faster workflows. Faster decisions. The promise is simple: remove the steps, automate the task, reduce the effort and let the system do more.
But the more AI becomes embedded into products, services and software, the more that promise needs challenging.
Because AI does not automatically make people more productive.
In many cases, it shifts the work. From doing to checking. From deciding to supervising. From creating to correcting. From acting to recovering when the system has misunderstood the task, misread the context or moved too far without the user’s consent.
That is the hidden productivity problem in many AI products.
Bad AI products do not remove friction. They move it downstream.
They make the first action feel effortless, then leave the user with the burden of verification, clean-up, rework, explanation and risk.
The next phase of AI product strategy cannot simply be about automation. It needs to be about controlled productivity: helping people move faster without losing judgement, confidence or control.
AI has moved from novelty to infrastructure.
AI is no longer a separate tool that people consciously choose to use. It is becoming part of the everyday systems through which people shop, search, work, book, learn, compare, complain, create and make decisions.
It is appearing inside retail journeys, customer service flows, productivity tools, workplace software, SaaS platforms, connected products and enterprise systems. Microsoft’s 2026 Work Trend Index frames this shift around agents, human agency and the redesign of work itself, based on Microsoft 365 productivity signals and a survey of 20,000 AI-using workers across 10 countries.
That shift matters because AI is moving from suggestion to action.
It is not just summarising information. It is recommending what to buy, drafting what to say, deciding what matters, automating workflows, triggering tasks and increasingly acting on behalf of the user.
For product and service teams, this changes the design challenge.
The question is no longer just, can the AI do the task and I’ll edit later?
The better question is: can people understand, steer, challenge, edit, approve and reverse what the AI is actually doing?
The productivity promise is under pressure.
Most AI product narratives assume that fewer steps means a better experience.
But productivity is not just speed. Productivity is the ability to make meaningful progress with less wasted effort, fewer mistakes and greater confidence.
An AI feature that generates a quick answer is not productive if the user then has to spend ten minutes checking whether it is right.
An AI shopping assistant is not productive if the recommendation feels opaque, biased or commercially pushed.
An AI customer service bot is not productive if it blocks the user from reaching a human.
An AI workplace tool is not productive if it creates more summaries, more notifications and more outputs that nobody fully trusts.
An AI agent is not productive if it acts faster than the organisation can govern.
This is where many AI products create a productivity tax.
The tax is paid in checking, correcting, rewriting, approving, undoing, escalating, explaining and rebuilding confidence.
For individual users, this feels like cognitive load.
For teams, it becomes operational drag.
For brands, it becomes a trust problem.
For businesses, it becomes a product adoption problem.
Consumers are interested, but they want boundaries.
The appetite for AI-enabled experiences is real. Capgemini’s 2026 consumer research found that 25% of consumers had already used generative AI shopping tools in 2025, with a further 31% planning to use them in the future. But the same research also found that 76% want clear rules for when an AI assistant acts, and 71% are concerned about how GenAI tools use their data.
That tension is important.
People are not rejecting AI. They are rejecting uncertainty.
They want the benefit of automation without feeling exposed, manipulated or bypassed. They want help, but they also want boundaries. They want relevance, but they also want transparency. They want speed, but not at the expense of control.
This is why AI trust is not just a compliance issue or a UX detail.
It is a product strategy issue.
If people do not understand what an AI system is doing, why it is doing it and what control they still have, they are less likely to trust the experience. And if they do not trust the experience, they are less likely to adopt it, return to it or recommend it.
Seamless is not always better.
For years, digital design has treated friction as the enemy.
Fewer clicks. Shorter forms. One-tap checkout. Invisible personalisation. Instant recommendations. Automated actions.
In many contexts, that has improved the user experience.
But AI changes the meaning of friction.
When a product is simply helping someone complete a clear, low-risk task, seamlessness can be useful. But when a system is interpreting context, making recommendations, generating outputs or acting on someone’s behalf, removing every step can remove the moments where trust is built.
A fully seamless AI experience can hide too much.
It can hide the source of a recommendation.
It can hide the assumptions behind an output.
It can hide how personal data is being used.
It can hide whether the system is confident or guessing.
It can hide when the user has consented to action.
It can hide who is accountable when something goes wrong.
In those moments, friction is not a failure of design. It is a form of protection.
The aim is not to make AI products slower. It is to make them more legible.
Designed friction is a productivity feature.
Designed friction means placing intentional moments of control into an experience.
Not clumsy friction. Not unnecessary steps. Not the bureaucratic friction of bad forms, dead-end chatbots or confusing permission screens.
Designed friction is useful friction.
It helps people understand what is happening, make better decisions, avoid mistakes and recover quickly when something goes wrong.
In AI products, that could mean:
A confirmation step before an AI assistant sends, buys, books or changes something.
A preview before generated content is published.
Editable outputs rather than locked responses.
Clear explanations for recommendations.
Visible sources, assumptions or confidence levels.
Undo, rollback and version history.
Human handover when the situation is emotional, complex or high-stakes.
Autonomy settings that let users decide how much the system can do.
Approval thresholds for sensitive actions.
Activity logs that show what an AI agent has done and why.
These are not just UX features. They are trust mechanisms. They are productivity mechanisms. They reduce the risk of rework, confusion, wasted time and user hesitation.
Gartner’s 2026 guidance on AI agents makes a similar point from an enterprise perspective: governance needs to account for an agent’s autonomy level and scope, rather than treating all AI agents the same.
That logic also applies to product experience.
The more autonomy an AI system has, the more carefully its control points need to be designed.
The best AI products will support judgement, not replace it.
The future of AI product strategy is not about making the user disappear.
It is about making the user more capable.
That distinction matters.
A weak AI product turns the user into a passive recipient. It produces an answer, hides the logic and expects acceptance.
A better AI product creates a shared decision space. It helps the user compare, refine, choose, challenge and act with greater confidence.
In retail, that might mean AI narrowing the options while the user defines the values: price, quality, style, sustainability, dietary needs, urgency or risk.
In SaaS, it might mean AI drafting a workflow while the user approves the logic, edits the sequence and sets the boundaries.
In customer support, it might mean AI resolving simple issues quickly, but making human escalation obvious and frictionless when the situation becomes complex.
In workplace tools, it might mean AI summarising a meeting, but clearly showing what was assumed, what needs action and what the user should verify.
In connected products, it might mean the system learning behaviour over time, but making those learned preferences visible, adjustable and reversible.
The goal is not passive automation.
The goal is human-machine co-authorship.
AI should help people think, decide and act better. It should not simply encourage them to accept whatever the system produces first.
From AI features to AI trust principles.
This is where many organisations need to slow down strategically.
Too many AI features are still being added because the technology is available, not because the role of AI in the user journey has been properly defined.
That creates shallow AI integration: a chatbot here, a summary tool there, a recommendation layer added on top of an existing journey.
But trustworthy AI experiences need more than features.
They need principles.
Product teams need to define:
Where should AI advise?
Where should AI act?
Where should AI ask for permission?
Where should AI explain itself?
Where should AI show uncertainty?
Where should the user be able to edit?
Where should the user be able to undo?
Where does a human need to come back into the loop?
Where could automation damage confidence?
Where does control create more value than speed?
The questions sit at the intersection of proposition development, journey mapping, service design, product strategy and brand trust.
Control is becoming a competitive advantage.
As more products add AI, the novelty will fade.
Simply saying “powered by AI” will not be enough. In fact, for many customers, it may become a reason to be cautious.
The products that win will be the ones that make AI feel useful, understandable and safe to rely on.
That does not mean making the AI timid. It means making its role clear.
A strong AI product should know when to act, when to ask, when to explain and when to hand back control.
It should reduce unnecessary effort while preserving meaningful judgement.
It should help users move faster without making them feel rushed.
It should create confidence, not dependency.
It should make automation feel adjustable, not imposed.
Control will become part of the value proposition.
For consumer brands, that means designing AI-enabled experiences that feel transparent, human and respectful.
For SaaS businesses, it means building workflows where AI saves time without creating hidden review burdens.
For service organisations, it means knowing which moments can be automated and which moments still require empathy, escalation or accountability.
For innovation teams, it means moving beyond AI experiments and designing experiences people can actually adopt.
The strategic question has changed.
The first wave of AI product thinking asked:
What can AI automate?
The next wave needs to ask:
Where should AI act, where should it advise, and where must the user stay in control?
That is the real design challenge.
Because the best AI products will not remove every moment of friction. They will use friction intelligently to create clarity, confidence and control.
They will not treat productivity as a race to remove the user. They will treat productivity as a way to make the user more capable.
AI can make products faster. But control is what will make them trusted. And trust is what will make them used.
Building AI products people can actually trust.
If you are building AI into a product, service or customer experience, the strategic question is not just what the technology can do.
It is what people need to trust it doing.
That means understanding the user journey, the moments of uncertainty, the decisions that require confidence, the data people are being asked to share and the points where automation needs clear boundaries.
This is where experience strategy, user journey design, proposition development and AI trust principles become essential.
The opportunity is not simply to add AI.
It is to design AI-enabled products and services that people understand, control and want to keep using.
If your team is exploring where AI should sit within a product, service or customer journey, now is the moment to define the role it should play. Before speed becomes rework, automation becomes confusion and innovation becomes another layer of friction.
