Blog: AI agents
AI agents are being given ongoing responsibilities
Published by Forever Foundry
On 29 September, OpenAI introduced “dots”: always-on AI agents designed to work towards goals across connected applications without requiring every step to be directed.
OpenAI describes agents that can monitor customer feedback, update proposals as requirements change and coordinate ongoing projects. It is also piloting specialist agents with their own identities, credentials and access to company systems for responsibilities including procurement, invoice processing and customer support. Read the launch coverage.
This is a product announcement and an early rollout, not evidence of measurable value for other organisations. OpenAI also acknowledges that the agents can make mistakes and that consequential work should be reviewed.
The FOFO™ view is that the important shift is organisational, not simply technical.
When an AI agent moves from completing a task to managing an ongoing responsibility, leadership must decide where that responsibility begins, where it ends and who remains accountable for the outcome.
Work rarely belongs to one department
Imagine a manufacturer using an agent to manage supplier exceptions.
Procurement may want it to find alternative stock and reduce costs. Operations needs continuity of production. Quality must ensure that substitute materials meet the required specification. Finance controls what can be spent. Customer service experiences the consequences if an order is delayed.
An agent designed from procurement’s perspective could perform its task perfectly and still create a problem elsewhere.
The real unit of work is not “find another supplier”.
It is “keep production moving without compromising quality, margin or the customer commitment”.
That outcome crosses several functions. Understanding it requires listening to the people who perform the work, approve exceptions and deal with the downstream consequences.
Only then can the organisation decide which parts an agent should handle, which decisions need human judgement and where accountability sits.
Redesign the responsibility before automating it
Before giving an AI agent ongoing work, define:
- The business outcome it is expected to improve
- The information and systems it genuinely needs
- The decisions it may make independently
- The circumstances that require human judgement
- The person accountable for the complete outcome
- The effect its actions could have on other teams
Measure the end-to-end result, not simply the volume of tasks completed.
For the supplier example, that could mean production interruptions avoided, time taken to resolve exceptions, additional purchasing cost, quality failures and orders delayed.
An agent processing more cases is not evidence of improvement if people elsewhere spend more time correcting its decisions. Read why your AI agent needs a spending limit.
The FOFO Process™ creates the organisational evidence needed before responsibilities are reassigned. AI Compass™ listens across functions, roles and levels to expose the complete problem. Slid3rs™ helps your people rank the opportunity against other demands for investment. FlowRoom™ brings leadership together to agree the outcome, ownership and roadmap.
The arrival of always-on agents does not make organisational understanding less important.
It makes partial understanding more dangerous.
Before asking whether an AI agent can own the work, establish who owns the result, and whether the whole organisation agrees on what a good result means.
If a supplier builds the agent, ownership needs designing in from the start. Read why the fastest AI build can create the longest dependency.
Understand the complete business challenge before automating it. Explore AI Compass™.
Related reading
Next