← Back to the blog

Blog

AI in Projects: More Data Is Not More Knowledge

Antonio Bassi sums it up neatly in the current GPM blog: more data does not mean more knowledge. In projects, AI amplifies whatever knowledge processes are already in place. Good ones get better, weak ones become visible faster. The post from 28 August condenses what Bassi lays out in depth in his original paper on “Human-AI Hybrid Knowledge Ecosystems” for project-based organisations, and it matches what I see every day in software and IT projects.

I use AI in every one of my mandates, for analyses, evaluations, documentation, and increasingly for building my own tools. Precisely because I am convinced of its potential, I can see where the real value sits: in the architecture that turns information into reliable project knowledge, not in producing ever more of it.

Where the paradox strikes in day-to-day project work

Four situations from day-to-day software delivery where more AI output does not automatically lead to better decisions:

  • AI status reports in the steering committee: The summary reads print-ready. But who vouches for the number if nobody has interpreted it? A report without an accountable sender is a statement by the model, not a basis for decisions.
  • Test and defect triage: AI classifies hundreds of defect reports within minutes. If no evaluation rules are defined up front, this creates false precision: the numbers look exact, yet nobody can say by which criteria they were prioritised.
  • Automatic summaries: Meeting minutes and document digests save time but lose context. The decision log, recording who decided what, when, and on what basis, remains the project’s knowledge anchor, and no summary can replace it.
  • Accountability: AI can analyse, but it cannot take responsibility. Every AI output that feeds into a project decision needs a human owner who has reviewed it and stands behind it.

Knowledge architecture instead of a tool list

Bassi calls this a knowledge architecture: clear rules on who validates which AI outputs, how decisions stay traceable, and where human judgement is mandatory. That matches the finding of PMI’s Pulse of the Profession 2026, which explicitly describes complexity as a system and leadership problem rather than a tooling problem. If you bring AI into a project without defining these rules, you end up with faster reports about the same fog.

What project leaders should do now

Three things before the next AI analysis lands in a steering meeting. First, name an owner for every AI output who validates it and takes responsibility. Second, define evaluation and validation rules before automating, not afterwards. Third, keep a decision log that records decisions together with their basis; it is the project’s memory, and it keeps paying off after go-live. What counts is not how many analyses the AI delivers, but how many decisions have become traceably better because of them.

← Back to the blog