Agent factories will not stay in IT
Software teams already work in a loop where agents prepare and people approve. Procurement and customer service are next. The first to benefit will have a written process.
- AI agents
- agent factory
- process
Blog
Notes on transformation that has to work: change capacity, metric definitions, AI pilots with a baseline, platform tests on your own data and resistance as information.
The pilot worked. Scaling is a different project
A pilot succeeds with one team, chosen data and the consultants in the room. Scaling meets other processes, worse data and nobody's budget. Treat it as a new project.
- pilots
- scaling
- rollout
The steering committee that steers nothing
A monthly committee receives status slides and approves everything, while the decisions that mattered are taken elsewhere or not at all. How to run one that steers.
- governance
- steering committee
- decisions
Android XR and on-device agents: a change, not a gadget
Agents that live in glasses and phones arrive as a platform decision. What to settle first: dependence on one vendor, rules for agents, trust, and a test without hardware.
- Android XR
- AI agents
- vendor lock-in
Should you pilot XR glasses with an AI agent?
Display glasses with a built-in AI assistant are now priced and shipping. Before a pilot, check whether the glasses change anything in the process, and at what cost.
- XR glasses
- AI agents
- pilot
Measure the baseline before the AI pilot starts
An AI pilot that cannot be compared with how the work was done before gets judged on impressions. Two weeks of measuring first costs less than arguing later.
- AI pilots
- measurement
- baseline
Resistance to change is usually a rational answer
When people resist a new system, the usual response is more communication. More often they have spotted a real cost the plan missed, and it pays to find out which.
- change management
- adoption
- resistance
Test the platform on your data, not the vendor's demo
Every platform looks right in a demo built for it. A short test on your own messy data, scored against criteria written beforehand, shows what the demo hides.
- vendor selection
- platforms
- proof of concept
Two departments, two revenues: definitions before dashboards
When sales and finance report different revenue for the same month, a better dashboard will not help. What helps is an agreed definition with one owner.
- data foundations
- BI
- metrics
Your people already use AI. Start the policy there
Before the AI strategy is signed off, staff are already pasting work into public chatbots. A ban drives it out of sight; a policy built on what they do works better.
- AI adoption
- AI policy
- data protection
The constraint is change capacity, not strategy
Most organisations know what they should do. What they do not have is the ability to absorb another change on top of the four already running, and nobody measures it.
- change management
- capacity
- programmes
A consultancy that resells cannot advise you
The advice may still be good. The problem is that you have no way to tell, because the same firm profits from one answer and not from the others.
- conflict of interest
- vendor selection
- procurement
AI agents in a company that cannot describe its own process
Automation applied to a defined process removes work. Applied to an undefined one it produces confident output nobody can check, faster than anyone can review it.
- AI agents
- process
- automation
Six weeks of working software beats six months of design
A design document cannot be wrong in any way that hurts yet. That is precisely what makes it a bad instrument for finding out whether an idea is any good.
- pilots
- delivery
- innovation lab
How to buy consulting so that it has to work
Most consulting contracts are structured so that the supplier is paid the same whether anything changes. Five things to put in the scope instead.
- contracts
- scoping
- incentives