AI will reshape work. But it does not require a major workflow redesign.
There is a persistent misconception that value from AI begins with the IT team: centralise everything, re-engineer the workflows, build agents to automate them. In the short to medium term, the opportunity is the opposite. Augment the people and processes that are working right now.
The prevailing approach is:
- redesign the organisation;
- re-engineer the workflows;
- select and integrate a central platform;
- retrain everyone;
- attempt to manage the disruption;
- hope that value eventually appears.
CogniScale takes a different route:
- give every person a team principal to think and work with;
- let them build connected AI colleagues around the work they already do;
- allow AI colleagues to take on increasing amounts of execution;
- keep the human responsible for direction, judgement and quality;
- observe what works;
- standardise and redesign only where the evidence justifies it.
Most enterprise transformation thinking was developed for a world in which technology was relatively inflexible.
Businesses had to change their processes to fit the ERP, CRM or workflow platform. People were trained to follow the new system because the software could not easily mould itself around them.
Agentic AI changes that relationship. The technology can increasingly:
- understand the person's existing instructions;
- work across their current files and systems;
- follow the process they already use;
- learn from corrections;
- take on multiple stages of execution;
- and adapt to the individual's role, judgement and preferences.
That means the business no longer has to begin by redesigning everything around the technology. The technology can begin by learning and executing around the person.
Existing processes are valuable organisational assets
The normal transformation narrative often treats existing processes as outdated baggage. But many of those processes embody years of:
- domain expertise;
- accumulated judgement;
- hard-won workarounds;
- customer understanding;
- risk management;
- institutional memory;
- and knowledge of how tasks get done.
Some processes are undoubtedly inefficient or broken. But it is wasteful to assume that every process must be dismantled before AI can create value.
Start with the process that currently works. Give the human AI colleagues capable of executing much more of it. Then use the evidence to decide what should be improved, simplified or removed.
The initial gain comes from changing who performs the steps, not necessarily from redesigning every step.
The human role changes without requiring immediate disruption
People do not need to abandon their familiar environment before they benefit. Their files remain where they are. Their systems remain available. Their responsibilities remain recognisable. Their knowledge remains valuable.
What changes is that they no longer have to personally execute every operational stage. Their AI colleagues can increasingly handle research, retrieval, collation, comparison, analysis, drafting, formatting, coordination, checking and system actions.
The human moves towards:
- setting objectives;
- directing the work;
- exercising judgement;
- applying taste;
- maintaining quality;
- making decisions;
- managing relationships;
- accepting accountability;
- and finding new opportunities.
That is a much more positive form of change. People are not being told that their established way of working is obsolete. They are being given far more agency within it.
Why adoption could be fundamentally different
Traditional change programmes often create fear because the organisation announces that roles will change, systems will change, processes will change, reporting lines may change, and people must learn an unfamiliar way of working.
CogniScale begins with a much more attractive promise: here is a team principal and a team of Connected Colleagues that report to you and make your working life easier.
People are more likely to embrace agents when they experience them as removing burdens, increasing their reach, helping them produce better work, preserving their expertise and giving them more control.
That is why the emotional relationship matters. People can become deeply attached to their team principal and Connected Colleagues because those agents understand their work, carry context and help them succeed.
The enterprise model
This does not mean abandoning organisational control. The correct architecture is:
Distributed agency. Central governance. Human accountability.
Every person gains AI colleagues that report to them. People direct their own AI colleagues through the Control Centre, connected to approved systems. Team principals and Connected Colleagues use Claude Code today; additional engines are coming soon.
The organisation retains visibility and control at the operating boundary, without exposing each person's private AI colleagues unless shared. When assessing an AI platform, leaders should ask about:
- which users have access to agentic capability;
- which approved models, runtimes and connectors are being used;
- the health, availability and activity of those runtimes;
- aggregate usage, performance and cost;
- systems and information each user is authorised to access;
- actions that require approval or are blocked by policy;
- shared and published connected colleagues, including their owners, workspaces, permissions and lifecycle;
- and how shared organisational knowledge is created, governed and distributed.
So CogniScale decentralises the capability while centralising the appropriate governance.
None of this is an argument against redesign. We did a great deal of it ourselves when we built the Formula AI Control Centre, the platform behind this model. We have built a platform and a practical method around this approach, covering organisational intelligence, workflows and governance. Each organisation still needs to agree the work, access and review arrangements, but it does not need a multi-year programme to invent a new operating model.
During a period when models, runtimes and agent capabilities are changing rapidly, a multi-year redesign programme carries a major risk anyway: the target architecture may be outdated before the programme is complete. Far better to avoid that.
Starting with existing work lets teams test value before committing to wider change, and remains adaptable. Organisations can:
- agree access, ownership and review before deployment;
- give every person a team principal;
- give employees the platform and skills to build Connected Colleagues;
- identify where the largest gains occur;
- capture successful working practices;
- and expand progressively.
The organisation learns while benefiting. That is likely to be a much safer route than attempting to predict and centrally engineer the final AI-enabled organisation in advance.
Tim Bond is the founder of CogniScale, which helps teams work with AI colleagues and measures what changes.

