AI is starting to change the shape of work

The AI news has felt particularly intense lately

I listen to The Artificial Intelligence Show pretty regularly, and the last couple of weeks have been a lot, even by AI standards.

OpenAI released GPT-6 Astra, with major advances in computer use and the ability to handle more complex professional work across applications. A 700-person bank shared how it is building AI-powered “digital employees” into its operations. There are increasingly serious conversations about what happens to entry-level jobs. Professional services firms are starting to face pressure from clients who know AI should make at least some work faster and less expensive. And at the same time, the safety conversation is getting more serious as the systems themselves become more capable.

At first glance, those can sound like separate stories. I don’t think they are.

For the last few years, much of the conversation about AI at work has focused on individual productivity. Write this faster. Summarize that meeting. Research this company. Analyze this spreadsheet. Help me brainstorm. Turn these notes into something usable.

Those uses still matter. I use AI this way constantly.

But I think the more consequential shift is happening one level above the task. We’re starting to ask what happens when AI changes the workflow itself.

Productivity and business impact are still two different things

McKinsey’s 2026 State of AI research puts some useful numbers behind that distinction.

High performers attribute at least 5% of EBIT to AI and report significant value from it. Half of all respondents also said AI helps them make better decisions.

That gap makes sense to me.

Saving 20 minutes writing an email is useful. So is getting a first draft faster or cutting an hour out of research. But a collection of personal productivity gains does not automatically add up to a meaningfully better business.

What stands out in the McKinsey research is the difference in ambition. The small group seeing stronger financial results is applying AI more broadly and using it as an opportunity to reconsider how the organization operates. Those companies are 3.3 times more likely than their peers to be planning significant business transformation around AI over the next three years.

That feels much closer to where the real opportunity is. Take a recurring piece of work from beginning to end.

Where does the information come from? Who touches it? What gets copied from one system to another? Where does someone have to wait for someone else? Which decisions actually require experience and judgment? Which steps exist mostly because that is how the process has always worked?

Then figure out where AI belongs. That exercise is much more valuable than adding another AI subscription.

A 700-person bank offers a surprisingly useful example

One of the recent episodes I found most interesting was The Artificial Intelligence Show’s conversation with Peapack Private Bank & Trust.

Peapack began experimenting with AI in 2023. By the end of that year, employees had access to AI chat and meeting-summary tools. The bank then created a network of 29 AI champions working across different parts of the organization.

What caught my attention was where the ideas started coming from.

Rather than relying on a technology team to continually introduce tools and suggest uses for them, people who understood the day-to-day work began identifying the problems AI might actually help solve. That shift matters.

Peapack has since used AI-assisted development to change its traditional build-versus-buy equation. In one example, it considered software that would have cost about $375,000 per year. Its internal team instead built a platform tailored to its account-review process, and the bank says those reviews now take about half as much time.

It has also introduced what it calls “digital employees.” One monitors IT support tickets. Another reviews the bank’s websites for things like outdated content and broken links.

The takeaway for me isn’t that every business now needs digital employees. It’s that the useful AI ideas came from understanding the work.

That should be encouraging for smaller companies.

A five-person marketing team can do that. So can a 30-person professional services firm. So can an owner-led business where everybody already knows the three processes that drive them crazy every week.

You may not have the technology resources of a big enterprise. You also probably don’t have as many layers between identifying a problem and changing the way something gets done. That can be an advantage.

AI is starting to challenge how professional services are priced

This is one of the developments I think B2B service businesses should pay particularly close attention to.

A recent Artificial Intelligence Show episode discussed pressure from banks on outside law firms to pass along some of the efficiencies they are gaining from AI. The underlying research makes the tension pretty clear.

Source: Thomson Reuters, 2026 Future of Professionals research.

Legal may be one of the clearest examples right now, but I don’t think this stays confined to law firms.

Accounting. Consulting. Marketing. Research. Design. Software development. Financial services.

Any business where value has historically been closely connected to hours worked should be thinking about this. If something used to take ten hours and now takes three, eventually the client is going to notice.

That does not necessarily mean the work should cost 70% less. It does mean service businesses need a much clearer answer to a basic question.

What are clients actually paying us for?

Maybe it is judgment. Experience. Knowing which questions to ask. Understanding the client’s business. Seeing patterns someone inside the organization may miss. Making decisions. Knowing when an AI-generated answer sounds perfectly reasonable but is strategically wrong.

Those things become more visible when execution gets faster.

I think that is going to force some healthy conversations about pricing, deliverables and what professional expertise is actually worth.

There’s also a people problem we haven’t figured out

Another recent episode raised a question I keep coming back to: if AI starts doing more entry-level work, how do entry-level employees eventually become experienced employees?

Historically, people learn by doing.

A junior employee researches. Drafts. Analyzes. Builds presentations. Cleans up spreadsheets. Takes notes. Sits in on calls. Makes mistakes. Watches someone more experienced fix those mistakes.

Eventually, they start seeing patterns. That is part of how judgment develops.

If AI absorbs a meaningful percentage of that early-career work, businesses are going to need another way to create experienced people.

Thomson Reuters’ research makes the problem more tangible. Seventy-one percent of respondents believe early-career professionals will still need structured support from experienced colleagues to develop judgment, and they expect it could take roughly a year longer for those employees to reach a level of independent judgment the organization trusts.

That creates an interesting tension.

AI can make a senior person faster. It can potentially reduce the need for some junior execution. But if companies simply eliminate junior work without redesigning how people learn, where do their future senior people come from?

For leaders, I think this means efficiency cannot be the only factor when deciding what AI should take over.

Some work probably should disappear. Some may be better handled by AI with a person reviewing it. And some may still be worth having an employee do, at least some of the time, because the process of doing it teaches them something they will need later.

Employee development may have to become much more intentional.

More capable AI also requires more thoughtful guardrails

The other part of this conversation is risk.

There is a meaningful difference between asking an AI tool to help draft an email and giving an AI agent permission to access systems, make changes, or take actions on behalf of your company. That difference is becoming more important as the technology improves.

OpenAI says GPT-6 Astra is its first broadly deployed model to reach the “Critical” level for cybersecurity capability under its Preparedness Framework. According to OpenAI, the capability increase led the company to strengthen protections around harmful cyber actions as well as its own internal development and deployment practices.

The September 15 episode of The Artificial Intelligence Show also focused heavily on the broader debate over AI safety, independent evaluation and how quickly frontier systems should continue advancing.

For the average B2B company, I don’t think the takeaway is “AI is too dangerous to use.” It is much more practical than that.

• Access matters
• Data matters
• Permissions matter
• Approval points matter
• Human review matters

If an AI system can take action, someone needs to know what it is allowed to do, which information it can touch, what happens when it gets something wrong, and where a person should step in.

That is part of workflow design now too.

What I’d be doing right now

For most lean B2B teams, I would spend less time trying to keep up with every new AI release and more time studying the work already happening inside the business.

Start with a few recurring workflows that consume meaningful time or regularly create bottlenecks. Map what actually happens today.

Look at the repetitive execution. The information people need. The systems involved. The handoffs. The places work gets stuck. The decisions that require judgment.

Then experiment. Maybe AI handles 20% of the workflow. Maybe it handles 70%. Maybe you discover that the process itself makes very little sense and needs to be redesigned before AI should touch it. All three outcomes teach you something useful.

I would also start getting more serious about company knowledge. Messaging frameworks. Client history. Good examples. Processes. Decision criteria. Research. Institutional knowledge that currently exists mostly inside someone’s head.

AI gets much more useful when it has good context. Companies that organize that context well are going to be in a much better position to build useful workflows around it.

The bigger question is changing

For a while, the AI question for businesses was mostly: how can we use this to work faster?

I think the more useful questions now are becoming:

• How should this work happen in the first place?
• What should people still own?
• Where does our expertise create value?
• How do we develop people when machines do more of the early work?
• What does a better version of this business look like if AI capabilities continue improving?

Most companies do not need to become AI companies. But I do think every company needs to understand how AI could change the work happening inside it.

That feels like a much more useful place to focus than chasing the latest tool.

There is still a lot we do not know about where all of this goes. But there is plenty we can start rethinking now.

This article was informed by recent episodes of The Artificial Intelligence Show and research from McKinsey & Company, Thomson Reuters, and OpenAI.