Adopting Generative AI

Introduction

2026 Foreword

In early 2025, I developed a program to help leaders adopt generative AI chat UIs, which required helping leaders see those UIs as general-purpose business tools that would transform all knowledge work jobs and teams. At the time, the idea wasn’t novel, but the actionable roadmap had value. As I write in mid-2026, advances in UX such as Claude Cowork have automated some of the recommended actions, and also some of the unrecommended actions! The underlying principles remain relevant.

Getting Oriented

Generative AI is hard to wrap your head around. To understand how it can affect you as a leader, do this exercise:

1. Write down a few day-to-day issues facing you right now
2. Consider each issue in turn:

• Does it involve process concerns?
• Is GAI involved, or could it be?

Here is a list you might have come up with:

1. Is it time to hold an offsite with the team?
2. Why do the latest financials make me nervous, even though they are clearly strong?
3. Is it a problem that everyone on my team is using — or not using — AI in their own way?

In this case, all of them involve process concerns, and all of them can be accelerated by GAI:

1. Thinking through why you might need an offsite and planning it if one is warranted
2. Examining your feelings about the financials and determining whether you need to investigate and take action
3. Surveying the work behavior on the team and devising a set of best practices to train them on

You need a program for integrating GAI with your team at every level — individual tasks, assignments, team changes, and organizational direction.

The Program

1. Teach everyone, especially yourself, direct interactive use of the tool for mundane knowledge tasks:

• researching and summarizing
• reviewing material you’ve drafted
• “hiring contractors” (see below)
• The Key Technique (see below)

2. Manage the changes in your teams as they adopt the new skills of direct GAI

• Use (or adopt) a formal ownership process, such as R.A.C.I.
• Track changes to work and feed them back into the ownership record

3. Explore the new possibilities GAI opens up for (or forces on!) your business as a whole

We’ll imagine GAI adoption at a small, simple business: MyBuildings. It supports property management companies that handle 10–50 rental units. Those companies often begin as side hustles which take off. Initially, the founders run their businesses with Excel and SMS, but as they add units, the volume of tenant communications and TODOs demands a dedicated tool. MyBuildings’ software-as-a-service platform is designed for their specific needs:

• Tracking maintenance requests, orders, and results
• Answering lease questions with precise attention to the contract and law
• Mediating conflict between tenants, such as noise disputes
• Resolving conflict with tenants, such as move-out disputes
• Supporting users with a web app for office work and a smartphone app for managers working on site at their properties — a classic 2015 Internet B2B setup

We’re going to consider some examples of how MyBuildings might adopt GAI. In each example, we’ll assume the change goes well. Then we’ll see how even success in adopting GAI raises complex challenges to work through. (If MyBuildings were a real company, there would of course be mistakes and bad luck to manage through.)

The Tool

GAI accelerates a wide range of tasks

Consider a mundane task for one of MyBuildings’ entry-level workers: a customer has opened a support ticket asking “How do I set up rent reminders to go out automatically five days in advance?” The work of replying has multiple steps:

1. Find the current documentation
2. Scan it for how to control the timing
3. Summarize the information, adding a link to the full documentation
4. Check the reply for house style and other details before sending

GAI can accelerate each of these four micro-tasks. It is:

• A general purpose multi-tool for automating and accelerating knowledge work …
• … that disrupts and reorganizes the work it automates

GAI accelerates a wide range of assignments

Let’s say the CEO of MyBuildings is testing all the new features in the smartphone app that customers have responded so positively to. They wonder if the app is getting cluttered. They could ask the Head of Design to look into it, but that colleague’s time is valuable. It will only take a minute or two to grab some screenshots and ask a GAI to prototype a few simplified designs. The Head of Design, given two weeks to focus on a prototype, would certainly outdo the GAI. But the tool is good enough to quickly test the general idea of simplification.

Or imagine that the CEO is visiting customers in Seattle, one of their best markets. They eye the hot Vancouver market, just miles away. What’s involved in doing business in Canada? They could give an attorney specializing in cross-border businesses a $50,000 retainer and wait two weeks for a report — or they could ask the tool, and get a decent answer in seconds.

The tool doesn’t just give you an answer much faster — given how often you’ll decide having a person do the assignment is too expensive and cumbersome, it’s infinitely faster (and infinitely cheaper). You should hire these “free contractors” every day. (See below for risks and rules to keep in mind when doing so.)

The Key Technique

Could anyone duplicate your successful business just by Googling for how to do so? If they could, the business wouldn’t be successful. You have knowledge, relationships, and capital which set you far ahead of competitors starting from scratch. You also have insights into what matters now, and questions you are trying to answer. You are the expert on your business and on what matters to you.

GAI, meanwhile, knows what the Internet knows. Asking it ad hoc questions will produce summaries of what the Internet knows.

The Key Technique is using the tool to deepen and extend your most important ideas and concerns. There are several tactics you can employ:

1. Have it ask you the questions. Consider planning a birthday party for a friend. You may want help getting ideas, but you want those ideas to suit your friend. The tool needs to gather information from you to target its suggestions.
2. Load information into it. If an interior designer’s recommendations for redoing your space have you intrigued but nervous, the tool is going to do a better job understanding your concerns if you upload pictures and the recommendations.
3. Offload information from your mind. When we are thinking through a large complex problem, just keeping track of details can be a challenge. A session with a GAI tool can be like a session with a whiteboard or your notebook, where you write and draw to explore possibilities — except the tool can notice connections and possibilities that you haven’t gotten to, nudge you when you’re drifting off task or have forgotten a key point, etc.

We’ll show how this technique helps solve important problems below.

Risks and Rules

Like any other powerful tool, GAI can do damage if used carelessly. It can:

• Confirm your perspective by default rather than evaluating it objectively (“sycophancy”)
• Produce long documents whose fluency may hide inaccuracies
• When integrated with other programs (aka “cowork”) and given access to untrusted data and the Internet, allow hackers to read your files

GAI also has two “meta”-risks:

1. You experience it via complex systems beyond the ability of most people to fully understand and control
2. Those systems, and especially the underlying GAI capabilities, change much more quickly than even many GAI professionals can keep up with

Start with these rules:

1. Work involving GAI should feel like real work, done well
2. Read GAI output critically, calibrating your effort to the importance of accuracy
3. Know the law and relevant procedures (i.e. your company’s rules)
4. Don’t let the tool have unsupervised access to private data unless you really know what you are doing
5. Learn from others’ adoption of the constant stream of brand-new GAI capabilities before trying them yourself

What does “feel like real work” mean in practice? If we are doing real work well, can we skim a possibly informative document quickly? Sure. Can we skim a crucial document? No. Can we present a document we haven’t written or read carefully as valuable output of our own? No.

The Team

As your team automates tasks and assignments with GAI, they’ll free up some time. You’ll need it. Change creates confusion and even conflict, which takes time and attention to manage. Consider a MyBuildings Customer Support representative receiving a ticket: ”How do I set up automatic rent reminders?” The rep owns:

• the quality of the response
• the customer relationship (Alex hates chit-chat, Beth expects it)
• the responsibility to notice patterns and decide when to escalate (“everyone is mad about the new feature”)

Crucially, this ownership is so obviously part of the job it is usually not named and tracked.

Ownership won’t own itself

A GAI tool drafts a response to the ticket, which the rep edits.

1. Who owns its quality: the rep, the tool team, or the head of support?
2. Who manages the customer relationship?
3. Who decides when to escalate?
4. What if the rep is a good rep, but a bad editor?

Automating one simple task raises four ownership issues. You need a process for evolving ownership as automation proceeds.

A light ownership process

The first step in tracking who owns which responsibilities is to track the responsibilities themselves. This effort can seem overwhelming. We just saw how a simple customer support task encloses a diverse set of responsibilities. Start with a manageable, high-level list for each job (perhaps culled from the description you advertised, the offer letter, or performance evaluation forms). GAI will probably force you to break tasks down, but you can do that as you go along.

Once you have the tasks, you can track ownership. Pick a framework such as R.A.C.I. or D.A.R.E., and set an update mechanism — reactively when work changes, or periodically (“Any new GAI uses last week?”). Circulate the updates, paying attention to the need to escalate (“Since Marketing now does its own data analysis, we need to sort out their cybersecurity responsibilities”). All of these details must be managed — you’ll need to decide who is “Responsible, Accountable, …” for the R.A.C.I. itself.

Running experiments

The results of automating a task or assignment with GAI will often be unknowable beforehand. Your team will be running experiments, a distinct skill which individuals will have or lack to varying degrees. Let’s imagine that MyBuildings customer support reps find the replies drafted by their GAI tool consistently solid, and rarely edit them. You push your Head of Support to remove the human from the loop, while of course monitoring how it goes. Questions will immediately arise:

• What do we monitor (customer satisfaction, response quality, time saved) and how do we measure it?
• Should we optimize for efficiency or risk reduction? A quick win, or learning?

Now imagine that the Head of Support has strong opinions about the right way to do the job. Maybe they were a strategic hire for you — expensive, but worth it because they considered themselves the authority on good support practices, and delivered. Their personality may be particularly unsuitable for the new task of managing experiments precisely because they were such a good fit for the job before you started adopting GAI.

That example raises the question of how GAI should affect who does what at your company, or works there at all. Your current employees perform tasks at human speed that GAI can do at computer speed. Their aptitude for and willingness to integrate GAI task automation into their efforts will vary. You may find that some of your best people resist automation the most, as it threatens the source of their perceived advantage. All that said, in many businesses, the employees are the company, embodying its abilities, contextual knowledge, and working relationships. The impact of demoralizing or superseding a member of the team can have a wide blast radius. How should you handle it? That depends on the details of your situation, so try the Key Technique. Describe your people management challenges to the tool, then have it help you come up with and test ideas for resolving them.

Managing interactions

Two seemingly independent changes can uncover awkward dependencies. Let’s say GAI is reading and responding to customer texts and emails. Part of the input form requires that the customer pick a topic area for the message, such as “renter concerns” or “technical support”. This helps the GAI focus its answer. You also showed your Head of Design one of the simplified UI designs the tool suggested. After some modifications, she rolled out a less complex app that no longer has users pick a topic area before sending their message. Your two GAI initiatives, which didn’t seem to have anything to do with each other, revealed a hidden connection that has become a conflict! What are you going to do?

Problems with no obvious decision criteria such as these are good candidates for some ideation with the Key Technique:

1. Start a chat by uploading documentation of MyBuildings’ existing business strategy and situation
2. Explain the two initiatives and how they conflict
3. Have the tool interview you to fill in gaps
4. Ask it to help you formulate possible resolutions

Together, you might come up with options such as:

1. Revisit the decision to eliminate the topic choice from the UI
2. Accept the loss of the topic choice and see how the GAI’s answers hold up
3. Have a new AI layer determine the topic

(In late 2024, I built a system using the third choice. It worked well.)

The Business

Before GAI arrived, MyBuildings had a straightforward customer acquisition path. As small property managers grew, they found themselves wasting time and money cobbling together their own software stack. MyBuildings directly addressed their pain, and the per-seat (really, per-smartphone) licensing model enabled customers to try MyBuildings at low cost, then expand as it proved out. Their most successful customers literally grew their businesses around MyBuildings, providing durable revenue.

Now mom-and-pop property managers hear about ChatGPT long before they hear about MyBuildings, and leverage it for growth, slowing the incoming sales funnel. Perhaps worse, the big established property managers who provide the bulk of MyBuildings’ income want to hear how new AI-driven features will reduce their own costs, not MyBuildings’ costs. MyBuildings is getting squeezed. The CEO needs an accelerated process for revising their business model.

Extend your mind

Consider the strategic challenge facing MyBuildings. The CEO can’t delegate the work of choosing a path, but they can accelerate that work by leveraging GAI. They follow this three-step process:

1. Load their current understanding into the tool
2. Go back-and-forth with the tool to develop a sharper, shared understanding
3. Have the tool generate scenarios ranging beyond what they could explore alone

Step one is simple. They describe the challenge: “Our sales funnel is drying up as small managers leverage AI directly. Meanwhile, our big customers want us to automate their work, not ours. …”

Step two requires that they coach the tool into being a good thinking partner: “I want help thinking through this strategic challenge. Ask me questions as needed, then generate some testable hypotheses for how we should guide the business through these challenges.” (Note that the second half of this prompt frames step three for the tool, which helps it do a good job on step two.)

When I pretended to be MyBuildings’ CEO with Claude Opus 4.6, here were some of its questions for me in Step 2:

• “What’s the core value proposition that has worked historically?”
• “Are there new entrants positioning as ‘AI-native’ property management tools?”
• “How does MyBuildings monetize currently: per seat, per unit managed, flat fee, …?”

Here is a selection of the testable hypotheses it generated:

H5: Legal Compliance & Liability

• “Property management has legal exposure (fair housing, habitability, lease enforcement). Auditable, compliant AI may be worth paying for vs. free ChatGPT”
• Test: Ask customers and prospects: “What concerns do you have about using ChatGPT for tenant communications?” Survey property management attorneys about AI liability concerns.

H6: Pricing Model Alignment

• “If AI reduces seats needed, per-seat pricing works against you. Per-unit pricing aligns incentives.”
• Test: Time-and-motion study with 3-5 customers. Where do their reps spend time now? What would they do with 30% faster response drafting?

H11: Market Segmentation by Complexity

• “Mixed-use properties, short-term rentals, commercial+residential, or regulated housing have needs ChatGPT can’t easily handle”
• Test: Segment your customer base by property complexity. Do higher-complexity customers have better retention and lower AI substitution risk?

H6 is pure gold: Claude identified a core plank of MyBuildings’ business model that is misaligned with the current business context. MyBuildings’ job is to reduce their customers’ operational costs. With per-seat pricing, features that reduce staffing are good for customers but bad for MyBuildings. With per-unit pricing, they remain good for customers but are no longer bad for MyBuildings. Now it can develop new functionality for its best customers, increasing their loyalty.

Moving forward

Spreadsheets were invented in 1979 and instantly recognized as a transformative business tool. It took over a decade for companies to reorganize their work and themselves to take advantage of financial processing. The work went so deep that it changed what being a businessperson meant. Financial analysis, once a specialized task, had become a mandatory skill. GAI automates the processing of verbal and visual information, just as spreadsheets automate the processing of numbers. It will take years to recreate how we work. The best time to start on that work was in 2024. The second-best time is now.