AI consultant Igor Rikalo argues that companies should shrink the role of their AI centres of excellence as adoption grows, leaving business teams accountable for results while central teams provide shared standards and tools. His practical proposal is a useful alternative to making one central office approve every AI idea.
Watch Desk analysis
What happened
In a 6 October Forbes Councils article, Rikalo says an AI centre of excellence should focus on capabilities that teams need in common: architecture standards, model evaluation, data-access policies, security and agent permissions. Business teams, he argues, should decide what to build and own the outcomes, because they understand the work and its trade-offs.
His suggestions include a central set of approved connections to core systems, testing tools, prompt and policy libraries, and cost dashboards. For agents, he recommends a registry recording each system’s purpose, business owner, data access, permitted actions, escalation rules and whether it is live or retired. He also argues that significant AI-supported decisions should leave a record of their assumptions, recommendations, human overrides, outcomes and costs. Read Rikalo’s article.
Why it matters
The proposal puts a practical question at the centre of AI governance: how can organisations let teams move quickly without making every group rebuild the same plumbing or leaving nobody responsible when a system misfires? A central team can offer reusable foundations and visibility without becoming the approval queue for every project.
Rikalo is setting out an operating model, not reporting a controlled comparison showing it outperforms alternatives. Even so, the details make the argument testable: track how much work business teams deliver, how often deployments reuse shared components, and how long ideas take to reach production.
Our read
This is a stronger job description for an AI centre of excellence than “please review every proposal until the calendar gives up”. Set the common rules and infrastructure centrally; put ownership of the actual decision and its consequences with the business team using the system. That split is only credible if teams can see what their AI costs, what it can do and who is meant to intervene.
Treat Rikalo’s model as a proposal to test, not a governance recipe guaranteed to work everywhere. Organisations should start with shared visibility and clear ownership, then check whether reuse rises and delivery gets faster without accountability becoming harder to find.
What to watch
- Whether business teams can use shared connections, testing and cost tools without building parallel versions.
- Whether agent registries record permissions, escalation routes and retired systems, not just names.
- Whether reuse increases and time to production falls while ownership of outcomes stays clear.
Discussion spark: Should AI governance be centralised around shared standards and tools, or should business teams have more freedom to build and manage their own systems?
Sources and evidence
- Why Your AI Center Of Excellence Should Get Smaller As AI Gets Bigger – Forbes (6 October 2026, 12:45 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.