Geordie co-founder Hanah Darley and Skillable’s Jessica Zeba-Snow argue that businesses need to measure AI agents by the work they do, not just by people’s estimates of time saved. Their five predictions for 2027 point towards a more detailed way to track costs, results and which agents deserve to stay.
Geordie Watch analysis
What happened
In an article published on Geordie’s site, Darley and Zeba-Snow say self-reported time savings give an incomplete picture, particularly when agents work in the background. They expect organisations to use records of agents’ actions, tools, costs and outputs as evidence of the work performed.
Their other predictions cover separate value measures for different roles and types of AI; treating an organisation’s agents as a portfolio, with decisions about what to expand or retire; sharing value data with the managers responsible for the work; and closer collaboration between security and AI teams.
Why it matters
These proposals offer a practical alternative to the familiar “hours saved” calculation. A coding agent and a customer-service assistant do different work, so comparing both against one organisation-wide average may obscure where either is useful. Tracking outputs alongside cost could give decision-makers a better basis for investment than adoption figures or self-reported wins alone.
There is a catch: the authors present predictions and an approach informed by Skillable’s experience, not independent evidence that these measurement systems will reliably identify business value. Even so, they put a useful question on the table: what evidence should count when a company decides whether an agent is earning its keep?
Our read
The strongest idea here is to measure the work, not merely the enthusiasm around it. Agent activity records could make previously invisible work easier to assess, but a trail of actions is not automatically proof of useful results. Pairing evidence of output with quality and cost measures is the more convincing test. “We saved time” can open the conversation; it should not be allowed to end it.
What to watch
- Whether organisations publish evidence linking agent activity to useful work, quality and cost.
- Whether measures differ by role and type of AI rather than leaning on one broad average.
- Whether teams review overlapping agents and retire those that do not justify their cost.
- Whether AI, security, finance and business teams can work from the same value data.
Discussion spark: Should businesses judge AI agents mainly by measurable work and quality, even if that is harder to quantify than time saved?
Sources and evidence
- Source update (17 November 2026, 00:00 UTC)
Independent WittyWires tracker for public updates about Geordie. Not affiliated with or endorsed by Geordie; this is not an official account.