Welcome to Insights
Thoughts,
strategies, and
stories.

Enterprise AI Insights

Enterprise AI and operations
How can AI make a business better at what it does? I write about how teams choose where to use it, share responsibility for results, and test what improves.
( 01 )
Research and practice

What you’ll find here

Frameworks

How stakeholder value, responsibilities, and evidence fit together

Workflow pilots

How to test AI in a workflow with a real customer and a measurable outcome

Measuring results

How to tell whether improvements last and others can repeat them
( 02 )

Articles

What’s new
A row of glass chess pieces
5 min read
Excerpt:
Use stakeholder value, two checks at every handoff, and evidence of results to assess whether AI improves how a business operates over time.
Read the article

More insights

What’s new
01 · Insights

AI Is Separating Execution Speed from Outcome Accountability

AI can accelerate execution without resolving who defines an acceptable result or takes responsibility for it. As speed becomes less constrained by human time, clear standards and handoffs become more important to operating capability.

Read the full article →
02 · Insights

From People Adapting to Tools to Tools Adapting to Verifiable Processes

AI creates the possibility of fitting tools to clearly defined work rather than reshaping work around a tool's output. Verifiable processes provide the outcomes, standards, interfaces and takeover conditions that make this possible.

Read the full article →
03 · Insights

The Proof of Capability Is Shifting from Demonstration to Reuse

When AI makes one-off outputs easier to produce, a successful demonstration becomes less conclusive evidence of capability. Independent reuse tests whether a method can be transferred rather than merely performed by its original designer.

Read the full article →

Short insights

Observations from everyday work, starting with standards, repeat execution and handoffs.

01 · Short insight

Why AI Drafts Can Take Longer to Review

Drafts arrive faster, but reviewers may have to check more assumptions and sources. Agree on what counts as acceptable, then compare review time across similar samples.

Read the full note →
02 · Short insight

The Two Questions That Get Mixed Up at a Handoff

A deliverable can be acceptable while the next step is still waiting. Record those states separately so the team knows what to fix.

Read the full note →
03 · Short insight

The Team Tracks Output. Did Anyone Downstream Use It?

Generation, acceptance, downstream use and operating outcomes leave different evidence. Start with one concrete use and trace it back.

Read the full note →
04 · Short insight

When Someone Else Can’t Do the Work, What Evidence Is Missing?

A manual can describe the steps while leaving out the judgments an experienced colleague makes. Independent repetition helps locate what still needs to be documented.

Read the full note →
05 · Short insight

What Teams Have to Do Once the Work Gets Faster

Faster tools make previously informal work visible: setting standards, separating handoff decisions and helping someone else repeat the work.

Read the full note →

Further observations

Three additional perspectives on expert dependency, useful records and pilot scope.

Notes · 01

Faster Tools Make Dependence on Experienced People Easier to See

A lot of work has always depended on experienced people. They know when to allow some flexibility and when to hold the line. They also know whom to approach to move things forward. There may be process documents, but getting the work done still depends on these people filling the gaps.

Once AI speeds up some of the execution, that dependence becomes more visible. Generation can be quick, but someone who knows the context still has to look it over, make a few changes or arrange access. When that person is away, the speed advantage shrinks.

That is not necessarily a bad thing. It helps the team see sooner which parts of the work have not yet become methods everyone can use. Rather than continuing to rely on experienced people to hold it all together, this is a chance to write down outcome standards, handoff conditions and takeover points a little more clearly, then ask someone else to try following them.

In doing so, you may find that some details once dismissed as impossible to explain can, in fact, be captured. Once they are, experienced people remain valuable, but they are no longer indispensable at every step. The organization gradually becomes less likely to stall because one person is temporarily unavailable.

There is no need to change every process at once. Try a piece of work that currently depends on particular people and involves frequent handoffs, and the difference will begin to show. Faster does not automatically mean more capable. But speed can expose what was previously vague and create a chance to make it clearer.

Notes · 02

Records Do Not Have to Be Extensive, but Different Things Need to Stay Separate

As soon as someone says we need to keep evidence, people tend to picture another pile of forms. Often, though, the important thing is to record different things separately, not to record a lot.

At a handoff, for example, two questions may be enough: has this version passed acceptance, and can the next stage start now? If it has passed but work still cannot start, is the reason access, people or something else? Writing those answers separately makes it clear later where a change is needed.

A record that lumps everything together as 'delayed' or 'a problem' is hard to use afterwards. Separate records, however short, can be more useful than a lengthy summary.

Notes · 03

Make One Workflow Work for the People Taking It On Before Expanding Further

When we try to change many processes at once, it is easy to make a small change everywhere and end up unable to explain the effect anywhere. A better starting point is one workflow with a clear outcome, someone who uses it and someone responsible for it. Get it to the point where handoffs are reliable and others can do the work too.

Once people can reliably take that workflow on, consider whether neighboring processes would benefit from a similar approach. There is then a concrete point of reference for expansion, rather than a simultaneous rollout that leaves no one sure which parts were useful.

Making reuse work in a small area first makes problems easier to see and useful methods easier to retain than testing across a broad area all at once.

A broad scope also makes the baseline harder to define. Record a comparable starting point before judging which change made a difference.

Want to test an idea on a real workflow?