The new tool arrives with the usual fanfare: a brisk email from the executive team, a cheery town hall, a licence waiting in everyone’s inbox. There is often a demonstration in which a chatbot summarises a document, produces a presentation or turns a vague prompt into a plausible plan. People clap. Someone says “game-changer”. Then everybody returns to the work they had yesterday—except now with another tab open.

This is how a great deal of workplace AI is being introduced: as an accessory to work rather than an intervention in it. The result is not transformation. It is tool sprawl with better branding. You can see it in the browser of almost any knowledge worker. There is the sanctioned AI assistant, the unofficial AI assistant, the transcription tool, the meeting-summary tool, the presentation generator, the research tool, and a clutch of smaller apps that appeared because somebody watched a LinkedIn video at 11.43pm. Each promises to save time. Together, they create a new administrative hobby: deciding which artificial intelligence should do which bit of your job.

This is not a minor implementation problem. It is the central mistake of the current AI moment. Organisations are treating AI adoption as a procurement exercise when it is, in fact, a question of work design, managerial practice, human confidence and institutional courage. The technology may be new. The organisational reflex is not. Buy software. Announce software. Train people briefly on software. Measure licences. Wonder why productivity has not behaved itself. The companies that get further will not be the ones with the most tools. They will be the ones willing to ask a more awkward question: what should work look like now that these tools exist?

01 –

The great AI contradiction

There is an odd mood in offices at the moment. Employees feel they should be using AI. Many are already using it. Yet the organisations employing them often have no clear view of what “using AI well” actually means. Microsoft’s 2026 Work Trend Index calls this the “Transformation Paradox”: workers can be ready to use AI while the institution around them is not ready to benefit from it. Its findings are hard to ignore. Organisational factors—including culture, manager support and talent practices—accounted for 67% of reported AI impact, compared with 32% attributed to individual mindset and behaviour.

A person can be excellent at prompting, evaluating outputs and automating small pieces of work. But if their manager treats experimentation as a distraction, their team has no agreed way to share what works, and their performance goals reward only volume and speed, their new capability remains private. It might make one person’s Tuesday easier. It does not become a better operating model.

The same research found that 65% of AI users fear falling behind if they do not adapt quickly, while only 13% report being rewarded for using and experimenting with AI in their jobs. This is a very modern workplace instruction: please transform your work, but do not disturb the work.

Source: Microsoft 2026 Work Trend Index

“Employees are told to be curious, but not to make mistakes. They are asked to become more productive, but not given permission to stop doing low-value tasks.”

– Suchetana Bauri

02 –

Tool sprawl is a management failure

Tool sprawl is usually described as a technology problem. The clutter begins with subscriptions, overlapping features, passwords, dashboards, notifications and procurement requests. All true. But the deeper problem is managerial. Every new tool is a claim on attention. It asks an employee to learn a new interface, develop a judgement about when to trust it, work out where its output belongs, and explain its use to colleagues who may not use the same thing.

Without clarity, AI accelerates production but not judgement. It creates more content, more versions and more opportunities for an error to travel quickly through an organisation. A tool can be intuitive and still be badly introduced. Good user experience lowers friction; it cannot decide which work should disappear, who owns the final judgement or how a team should handle risk. A useful test is simple: if you removed the AI tool tomorrow, would the team still understand how the work gets done? If the answer is no, you have not redesigned a workflow. You have outsourced comprehension.

03 –

The false comfort of the pilot

Most organisations now have a collection of AI pilots. Pilots are attractive because they feel like action without requiring commitment. You can select an enthusiastic group, give them access, collect a few success stories and declare that the organisation is “on its AI journey”. But pilots often become a kind of corporate terrarium: a small, carefully tended environment in which new behaviours survive because they have been protected from the conditions outside. This is why successful pilots do not automatically scale. Scaling is not the act of giving more people licences. It is the act of changing the system that surrounds the licence.

WHAT PEOPLE NEED TO MOVE FROM PILOT TO PRACTICE:

• Understanding what is changing and why it matters to their specific work.

• What good looks like, what they are allowed to do, and where to ask for help.

• How their judgement still counts—identifying work that now needs more human attention.

04 –

Digital fluency is not prompt literacy

Prompting has become the office equivalent of touch-typing: useful, occasionally impressive, and not remotely sufficient as a measure of professional capability. Digital fluency means being able to judge what a tool is for, what it cannot reliably do, how its output should be checked, and when not to use it. It requires an understanding of how information moves through a process—and whether an apparently helpful shortcut creates a new risk somewhere else.

It also means moving beyond generic “AI literacy” programmes. People do not need identical skills because they do not do identical jobs. A legal team needs to understand review, traceability and risk. A customer-service team needs escalation rules and confidence about when empathy cannot be automated. A communications team needs editorial standards, source checking and a clear view of authorship. Managers need to know how to evaluate AI-supported work without rewarding superficial speed.

“A better approach begins with live work. Take a recurring task that people genuinely dislike and redesign it with the team. Set guardrails. Test outputs. Compare quality, time and risk. That is not merely training. It is organisational learning.”

– Suchetana Bauri

05 –

Workflow redesign: the part nobody can delegate

The most useful question a leader can ask is not, “Where can we use AI?” It is, “Where does our work currently waste human attention?” Look for repetition, delay, handoffs, avoidable rework, unnecessary reporting, duplication and decisions that wait in someone’s inbox because nobody is sure who has authority. AI can help with some of these. But it also exposes them. A messy workflow does not become elegant because a chatbot has entered the room. It becomes a messy workflow that produces more text.

WORKFLOW REDESIGN REQUIRES CHOICES:

06 –

Readiness is not a mood

Many organisations say they are “ready for AI” because they have bought a platform, appointed a steering committee or run an executive workshop. These are signs of activity. They are not proof of readiness. Readiness is the capacity to convert a new capability into better work at scale. Only 19% of surveyed AI users sit in the “Frontier” group, where individual capability and organisational readiness are both high. The figure should unsettle anyone treating AI deployment as an inevitable march from purchase order to productivity.

The bold claim is this: stop treating AI as a software category. It is a test of whether your organisation can learn in public. The harder test is whether leaders can admit what they do not know, managers can make room for experimentation without demanding instant proof, and teams can redesign work rather than merely speeding it up.Give people a tool and you may get a burst of activity. Give them clearer work, capable managers, practical guardrails and permission to learn, and you may get something rarer: a change that lasts.

07 –

Measure the change, not the logins

A licence count is not adoption. A login is not capability. A reduction in time spent drafting is not automatically a better outcome if it produces errors, confusion or work that nobody trusts. This is where many AI programmes become vulnerable to theatre. A dashboard shows that usage is rising. Leaders feel reassured. Meanwhile, employees have quietly built parallel processes to check the tool’s work, and the organisation is producing more material without becoming more effective.

What counts as adoption?

Measure at several levels instead: Access—who has the right tools and approved use cases. Usage—are people returning for meaningful tasks? Capability—can they use it safely in their roles? Behaviour—have routines, decisions and collaboration patterns changed? Quality—is the work more accurate, useful and trustworthy? Outcome—has the organisation improved a customer experience, reduced delay or made better decisions? Pair the numbers with listening. Ask employees what has genuinely improved, what has become harder, and what feels risky.

“An organisation that cannot hear these stories will misread its own data.”

– Suchetana Bauri

08 –

The real adoption strategy

The point is not to make every employee an AI enthusiast. People are allowed to be sceptical, cautious or simply busy. The point is to build a workplace in which useful experimentation is possible, boundaries are clear, good ideas travel and nobody is expected to figure out an organisational transformation alone from a prompt box.

Give people a tool and you may get a burst of activity. Give them clearer work, capable managers, practical guardrails and permission to learn, and you may get something rarer: a change that lasts. The next task is not to add more AI. It is to decide what better work requires—and be disciplined enough to remove everything that gets in its way.

The real test is broader: leaders must be able to admit what they do not know; managers need to create room for experimentation without demanding instant proof; and teams must redesign work rather than merely speed it up. Governance should protect people without making useful action impossible. Above all, organisations need to stop mistaking effort for progress and ask whether the work itself has become better.

“Stop treating AI as a software category. It is a test of whether your organisation can learn in public.”

– Suchetana Bauri

About Suchetana Bauri

Suchetana is an AI enablement consultant and communications strategist. She helps organisations make AI adoption understandable, practical and human-centred.

1. Microsoft, “2026 Work Trend Index: Agents, human agency, and the opportunity for every organisation.” Survey of 20,000 AI-using knowledge workers across 10 markets, February–April 2026.
2. McKinsey, “How to close the agentic adoption gap.” Identifies change management and siloed working as key obstacles to scaling agentic AI.