The Workforce Lens

What responsible use looks like
Responsible AI use is not just about choosing a good system.

It is also about making sure someone can question the result, understand the route that produced it, and stop the process when the stakes are too high for automatic handling.
That sounds obvious. In practice, organisations often rely on a job title, a policy document, or a final approval box and assume the work is done.

The real test is more concrete. Can the person responsible see enough, know enough, and act early enough to make a difference?

That is the question running through my article on AI agents and accountability.

Read it here: https://zurl.co/wK28R

AI Agent Accountability: Who Is Actually Liable When Something Goes Wrong New liability rules assign responsibility for AI failures while leaving the harder question unresolved

The practical test for accountability

Here is a simple test.
If something goes wrong, can the person named as responsible explain what the system was told, what it saw, what it did, and where they could have stopped it?

If not, the organisation may have assigned responsibility without assigning real control.
That distinction is easy to miss when a process looks neat on paper. An org chart can be clear while the decision path underneath it is not.

My article explores why this matters for AI systems, but the question applies more widely. Any organisation that delegates a consequential decision needs a record that people can actually use.

Read it here: https://zurl.co/WOxfi

AI Agent Accountability: Who Is Actually Liable When Something Goes Wrong New liability rules assign responsibility for AI failures while leaving the harder question unresolved

Labels change how carefully we look

A small change in language can alter how people behave.
If a piece of work comes from a tool, we tend to inspect it as something we are using. If it comes from an “employee”, we may start to treat it as someone else’s contribution, with someone else carrying part of the responsibility.

That is why the study discussed in my article matters. Managers caught fewer errors when the same work was framed as coming from an AI employee rather than a tool.
The concern is not the word itself. It is the shift in attention that can follow it.
When an organisation gives an AI a human role, it needs to be very clear about which human duties have not moved anywhere.

Read it here: https://zurl.co/4Cn5F

AI Agent Accountability: Who Is Actually Liable When Something Goes Wrong New liability rules assign responsibility for AI failures while leaving the harder question unresolved

A name on the chart is not control

Putting an AI agent on an org chart may help people understand where it sits in a process.

It may also create a false sense that someone has given it a proper role and someone else has proper control over it.
Those are different things.

A real role comes with authority, limits, supervision, and a way to challenge decisions. If the person listed as responsible cannot see the agent’s instructions or stop its actions, the chart is giving us a description without much control behind it.

The article asks what organisations are really saying when they call an AI a worker, an employee, or a boss.

Read it here: https://zurl.co/QBRUt

AI Agent Accountability: Who Is Actually Liable When Something Goes Wrong New liability rules assign responsibility for AI failures while leaving the harder question unresolved

This problem did not start with robots

Organisations have always found it difficult to assign responsibility when a decision is spread across departments, committees, and layers of approval.

AI makes the difficulty more visible because the chain can include systems that pass work to one another without anyone having a complete view of the process.
The basic human problem remains the same: people are expected to answer for decisions they may not have been able to see or change.

That is why the article looks back at corporate law as well as forward to AI regulation. The history is not a side note. It shows what happens when an accountability model is applied to a structure it was not designed for.

Read it here: https://zurl.co/bznQT

AI Agent Accountability: Who Is Actually Liable When Something Goes Wrong New liability rules assign responsibility for AI failures while leaving the harder question unresolved

Suppose the decision cannot be explained
An AI system rejects a loan.

The employee who reviews the case did not build the system. The manager who approved its use did not see the final recommendation. Somewhere in the process, another system altered the information that came through.
The customer asks why the decision was made.
Who can answer?

That is the part of AI accountability that interests me most. Not only who gets blamed when something goes wrong, but whether anyone can still explain what happened in a way that is honest and useful.

The new article follows that question through law, corporate governance, and the way organisations are beginning to describe AI at work.

Read it here: https://zurl.co/Tgokq

AI Agent Accountability: Who Is Actually Liable When Something Goes Wrong New liability rules assign responsibility for AI failures while leaving the harder question unresolved

Every AI liability law so far does the same thing: it decides who to blame without deciding what blame actually requires.

California blocks one legal defence. The EU treats AI like a product. Singapore says outright that AI agents are not legal persons. Different tools, same move underneath, push responsibility onto a human by default, because the harder question is not answered yet.

Traced that same gap back further than AI. Decades ago, corporate law hit the identical wall, on people, not machines, and a major fraud case collapsed for basically the same reason an AI agent escapes accountability today.

➡️ In my latest piece:
👉 Why the standard "AI cannot suffer, so it cannot be accountable" argument is weaker than it looks
👉 The counterargument some legal scholars now take seriously, that AI may eventually need something like legal personhood
👉 What a real study found happens to human vigilance the moment work gets relabelled "AI employee" instead of "tool"
👉 The old corporate law precedent showing this gap is not new, and was never really about AI at all
👉 Why the piece ends without resolving its own argument, and what that leaves every organisation still needing to work out

https://zurl.co/CVoxP

AI Agent Accountability: Who Is Actually Liable When Something Goes Wrong New liability rules assign responsibility for AI failures while leaving the harder question unresolved

Do not make endurance the business model

Some managers appear to thrive in player-coach roles because they are carrying an unreasonable load with remarkable discipline.

That can conceal the design problem.

The work gets done. The meetings happen. The team receives enough attention to avoid an obvious breakdown. Meanwhile, the manager’s own recovery time, thinking time, and capacity for difficult conversations disappear.

A role should not be judged sustainable simply because a strong person has managed to survive it.

The better test is whether the arrangement leaves enough room for quality technical work and quality management without requiring constant personal sacrifice.

https://zurl.co/WJHMe

Span of Control, Pay Gaps, and DRIs: The Hidden Limits of Modern Org Design Why dual career ladders, ownership roles, and player coach managers work in theory

The label often arrives before the fix

A company announces a dual career ladder. A team introduces DRIs. A manager is asked to become a player-coach.

The announcement creates a sense of movement. People have a name for the new arrangement and a diagram to share.

The harder part begins after that. Specialists want to know whether progression and pay are genuinely comparable. DRIs need to know what they can decide. Player-coaches need to know which work is coming off their plate.

Those questions determine whether the change reaches daily life. Without answers, the role may be easy to describe and difficult to perform.

The new article looks at those gaps, using research on pay, delegation, hierarchy, span of control, and manager workload.

https://zurl.co/PR2yv

Span of Control, Pay Gaps, and DRIs: The Hidden Limits of Modern Org Design Why dual career ladders, ownership roles, and player coach managers work in theory

A flatter chart can create a different workforce

When a company removes layers, the official reason is usually speed. Approvals should be quicker. Decisions should sit closer to the work. Senior leaders should have fewer steps between them and the people doing the job.

Those gains may be real. The change also alters the daily experience of employment.
People who are comfortable with ambiguity may stay. People who relied on a layer for coaching, context, or protection may decide that the new arrangement is not workable. Over time, the workforce becomes shaped by those choices.

The chart looks leaner. The population inside it has changed as well.
That deserves attention before “flat” becomes shorthand for “better”.

https://zurl.co/0HkUa

Span of Control, Pay Gaps, and DRIs: The Hidden Limits of Modern Org Design Why dual career ladders, ownership roles, and player coach managers work in theory