1. AI Agents Won’t Be Confined to a Web Page

My 11-year-old likes to Google news stories about ridiculous AI use cases, just to make fun of them:

“Hey Dad!  Did you know they have AI toothbrushes?”

“Hey Dad! Did you know you can buy an AI dog food bowl?”

Statements like these make me feel stupid saying “In 2031, AI will be everywhere,” but…

In 2031, AI will be everywhere.

Not necessarily “in your toothbrush” everywhere, but everywhere that matters² . And that will drive a structural change in how expertise is delivered.

Historically, professional expertise has been delivered in discrete engagements: a consultant visits, diagnoses a problem, runs a workshop and leaves, and the client is expected to remember and apply what they learned. As AI moves off the web page and into the phones, laptops, vehicles, equipment terminals, and enterprise software people already use, a consulting firm’s expertise can follow its clients to the point of performance. Instead of teaching a maintenance technician how to troubleshoot a machine and hoping it sticks, a firm’s AI practitioner can stand beside that technician (figuratively speaking) while they do the work.

To give a specific example, our company builds AI assistants for social workers and AI coaches for financial advisors. And while you might think these professions couldn’t be more different (one helps people overcome barriers to employment, the other helps millionaires invest their fortunes), they have one thing in common: they both spend a lot of time in their cars, driving around to meet clients.

After hearing customers from both groups say “I wish I could talk to [the AI] in my car, because that’s when I have time…” we updated our AI platform with Bluetooth steering wheel controls for hands-free operation.

But then another organization saw the ad for our steering wheel features and said, “Hey, could your AI integrate with a medical device?” So I went back out to my car and did a demo of the AI receiving telemetry from its diagnostic sensors, just as it would from a medical device.

The real irony, however, is that the AI agent in the driving demo isn’t my actual “driving assistant” AI, but rather my sales assistant AI, Max, that helped edit the script: which is as good an example as any of how AI won’t just stay on a website we view from our desks, but will follow us around from place to place, wherever we do our work.

This is a reality that Google in particular is aggressively pushing toward, working its Gemini AI model into its entire product suite, and it’s where the two halves of my own company’s business converge. Half of our work comes from helping expert consulting firms create agents that embody their knowledge, while the other half comes from building enterprise AI infrastructure, the boring plumbing that will pipe those agents wherever they need to go, with all the requisite security and access controls. And once those pipes reach everywhere, expertise stops being an event and becomes a presence.

That changes almost everything about the consulting business. Once people can get answers to their questions wherever and whenever they arise, from the nearest available speaker, booking an appointment with a human advisor will start to seem like an inconvenient extravagance in 99% of situations.

Which raises two much bigger questions: whose expertise is coming out of that speaker, and what becomes of the human experts behind it?

2. Orgs Will Want Outside AI for the Same Reasons They Hire Outside Consultants

Why do organizations hire consultants when they already employ so many smart people?

Sometimes it’s because the consultant has unique expertise, but other times it’s because the consultant doesn’t have a direct stake in the interpersonal dramas that drive organizational decision-making.

It’s not hard to imagine that plenty of organizations in 2031 will have dubious homegrown AI systems that exist because the wrong person got awarded a budget back in 2028, and now the AI system is a digital albatross that nobody listens to, or that butts into people’s work to offer unhelpful guidance at inconvenient times.

Alongside those, they’ll have the general-purpose assistants built into their enterprise software, which many will suspect of being pliant yes-men designed to keep the customer happy. Next to either option, an outside AI practitioner with no stake in the office politics, telling people what’s what, with its own principles and point of view, might seem highly appealing.

That level of independence and professional integrity can’t be achieved with a “personality” setting on a generic AI chatbot. It has to be baked into every aspect of the AI system’s design (the methodologies it follows, the knowledge sources it draws upon, the actual user experience of whatever platform or platforms it runs on) and earn a reputation for reliability and performance over time. There’s a difference between a generalist AI chatbot that acts “independent” by performing disagreement and one with a coherent methodology, evaluation criteria, and professional style, developed and maintained by a firm with specific domain expertise and a professional reputation on the line.

For instance, our team built an AI coach to help financial advisors build their clientele. One of the things it advises users to do is focus on a specific niche of affluent investors, like business owners or retirees. At one point while testing it, I pitched what I actually thought was a good idea:

Me: “Hey, I used to work at law firms [true] and know a lot of attorneys. Maybe I should market my services to lawyers!”

AI Coach: “That’s interesting… what specific kinds of lawyers do you know?”

Me: “Oh, lots of them. I know a small business lawyer, a real estate attorney, a litigator who does class action suits, an immigration lawyer…”

AI Coach: “Hmm… that sounds more like a professional network than a coherent market. An attorney who earns fees from high-profile lawsuits and someone who does transactional immigration work will probably have completely different income levels and approaches to investing. Maybe you could have the small business attorney introduce you to successful business owners?”

Contrast that with this actual conversation with generic ChatGPT:

Me: “I’m just getting started as a financial advisor, and I used to work at law firms [true] and know a lot of attorneys. Maybe I should market my services to lawyers.”

Generic GPT: “I think you may have stumbled onto a very smart niche for your financial advisory practice. And not just because lawyers tend to earn good money. Your real advantage is that you’ve worked inside law firms. You understand the culture, the personalities, the pressures, and probably some of the financial challenges attorneys face. That’s a much stronger starting position than being a brand-new financial advisor trying to convince random wealthy people to trust you. My recommendation: build a financial advisory practice specifically for attorneys, starting with the lawyers and law firms you already know.”

The fact that the specialist AI told me my idea was kind of stupid, based on the framework we’d given it, made that answer feel a lot more credible than ChatGPT’s enthusiastic “you may have stumbled onto a very smart niche!”

But this brings up a larger question every expert firm will have to answer: if a client doesn’t take kindly to your robot pushing back on their ideas and demands adjustments (as plenty of clients do with human consultants), will you have your AI agent’s back, and stand behind it the way you (hopefully) stand behind your human colleagues?

The idea of taking responsibility for an AI system might sound risky, but it could also act as a competitive shield for expert consultants. While big tech companies like Microsoft, Google, Amazon, and Oracle will continue to improve their AI products, it’s unlikely they will ever stand behind the output of a system within a particular domain. And why should they, when their real money comes from providing generic infrastructure to millions of organizations? What reason would they have to get deep in the weeds of ambulatory surgical center inspections or DWV engineering?³

A specialist firm, on the other hand, would be in a position to carefully and thoroughly evaluate an AI system’s fitness for a specific purpose, to the point where any risks become reasonable and insurable. And once AI agents can offer that level of reliability and accountability, robots can start building credible professional reputations.

3. Consulting Firms Will Reorganize Around Developing “Professional Intelligence”

If you want a preview of what’s about to happen to consulting firms, look at what’s already happened to software teams.

For decades, it would take a system architect a week to sketch out a new product feature, then a month or two for programmers to build it. Today, my company’s developers can code like lightning, but we can’t write specifications fast enough to keep the pipeline full.

The obvious question is “Why not have AI write the specifications?” Our architects do use AI as a sounding board. But while today’s flagship models can do a remarkable job of scanning a company’s codebase and seeing how the features fit together, they don’t come pre-installed with a specific perspective on how software should be designed, or with domain-specific discovery frameworks to ensure they are building in a way that will account for the organization’s future plans and ask the right questions about the larger business case. Thus, left to their own devices, even the best AI coding copilot will leave conceptual gaps big enough to drive a Cybertruck through, which is exactly how organizations end up in jams where they have to rehire human developers to debug. 

Put another way, AI is spectacularly good at knocking down dominoes, but less good at deciding where to set them up, especially when it needs to think beyond what’s been done before. That’s not a knock on the technology, just a description of the relationship. If something can knock down dominoes a thousand times faster than a human can, the human’s job becomes setting up better dominoes. Which is why we tell organizations to hire more architects and start thinking of coding as more of a quality assurance and troubleshooting role.

Consulting is about to go through the same inversion. Applying a framework to produce research, analysis, reports, and slide decks gets dramatically cheaper and faster. Deciding what ought to be produced, framing the problem correctly, and establishing what “good” looks like does not.

The firm of the near future looks less like a pyramid of billable consultants, with juniors churning out documents and seniors handling strategy and client interaction, and more like what I’d call a “professional intelligence laboratory”: a team of domain experts constantly challenging and refining the organization’s methodologies while conducting genuinely novel research to advance their field.

The lab’s core job is stewardship of the firm’s branded AI agents: ensuring their outputs align with the firm’s overall perspective while constantly integrating a flood of new information from clients, colleagues, and the wider profession, so the agents neither cling to outdated ideas nor get diluted into spewing bland conventional wisdom like a generic chatbot. Agents need to remain representatives of the firm, not just part of the tech infrastructure: “Isaac the AI consultant” has to remain Isaac, whether he’s running on the consultancy’s website, inside a client’s Microsoft environment, or integrated into a driver’s navigation system in a truck cab.

And just like human consultants, firms will need to walk a fine line between helping clients internalize their methods and frameworks (and now AI systems) and rendering themselves irrelevant.

Ever since the first astrologer advised a king on when to plant crops, clients have tried to learn the magic so they can stop paying the magician. Organizations might say, “Instead of charging $50,000 for 10 workshops, how about $12,000 to train our senior staff and leave behind your training materials? Then our senior staff can train the others.”

Whether or not consultants should accept such arrangements depends on a number of factors, but it’s worth noting that AI offers a new vector for the same dynamic. Every successful consultant knows how it feels to be pumped for free information. (I once had a $50 billion company schedule a “sales call” where 30 of their in-house team members showed up, and the first question was “Explain your methodology and how your technology works…” It made for an interesting discussion.)

Likewise, if clients have access to a firm’s AI system, they could act in bad faith and pump it for information rather than using it for its intended purpose. They could even use their own AI systems to accelerate the process. And this isn’t some hypothetical cyberpunk sci-fi scenario: in February 2026, Anthropic disclosed that three Chinese AI labs, DeepSeek, Moonshot AI, and MiniMax, had used roughly 24,000 fraudulent accounts to generate more than 16 million exchanges with Claude in order to train their own models on its outputs (a maneuver some would call “unethical” and others would call “clever”).

The bad news is you don’t need a futurist to tell you competition in the AI age will be as cutthroat as ever. Across the various consultancies I’ve worked for, we’ve regularly had situations where clients handed our IP to cut-rate competitors and encouraged them to rip off our firm’s approach. Likewise, AI distillation attempts will be more or less inevitable once firms start embedding their systems in client organizations. (we jokingly compare it to the old Lucky Charms cereal commercials, where a stereotypical cartoon leprechaun runs from a pack of hungry kids, shouting, “They’re always after me Lucky Charms!”) 

Fortunately, the good news is that there are multiple technical and legal ways to defend against it, including contractual clauses that allow the firm to shut down access if its platform detects suspicious activity. But distillation is one reason to think twice before naively agreeing to embed your AI systems inside a client-controlled Microsoft or Google environment: doing so isn’t always the wrong choice, and some clients will say “We want your AI running on our environment, take it or leave it…”, but it’s a risk firms should take with open eyes.

All that said, we sincerely believe that in the majority of cases, where everyone acts in good faith, providing clients with on-demand access to continually improving, intentionally designed, and carefully curated professional intelligence will create a deeper and more mutually beneficial relationship between consultancies and their clients.

Which brings us to the next prediction…

4. Clients Will Form Enduring Relationships with AI Advisors

As we mentioned earlier, the most obvious advantage of an AI consultant is availability. While human advisors require appointments, Isaac the AI can be there at 2 a.m., on the shop floor, in the car between meetings, or wherever somebody has a problem that needs solving.

But availability is only the beginning: the deeper advantage is continuity.

Imagine Isaac has worked with the same organization for four years. He knows its history, its priorities, the decisions that worked, the ones that didn’t, and the problems that keep coming back despite everyone’s promises to fix them. He remembers a conversation with a regional manager from eighteen months earlier and recognizes that the problem they’re describing today is suspiciously similar.

That doesn’t mean Isaac possesses magical insight, or that every recommendation will be correct. It means he’s combining the consulting firm’s professional methodology with an accumulating understanding of this particular client.

This is a dynamic I’ve already run into with my own AI advisors, Weston and Max. As I’ve described in other posts, Weston and Max are basically the digital angel and devil on my shoulders: Weston takes a more cautious approach than I normally would, while Max encourages bold action in cases where I tend to hesitate.

A few weeks ago, I described a business problem I was having to Weston, and his response was “Yes… and I have to say this seems to be a pattern for you. Remember when X? And the other time with Y? Same thing.”

Somehow, having a machine point out my own self-sabotaging tendencies landed harder than if it had come from a human. I couldn’t shake it off by saying “Weston just doesn’t understand our business model” or “Weston doesn’t have all the details of the situation,” when Weston has hovered over my shoulder while I drafted client emails and our business model is explicitly encoded into his knowledge base.

And while I’ve considered installing a “mute” button on Weston’s memories to prevent him from bringing up certain topics again, part of me hesitates, lest in doing so we undermine the independent perspective our team has come to rely on.

Of course, this kind of trust doesn’t come automatically. When we first began developing AI systems for clients, I liked to say, “A well-designed AI agent shouldn’t require a user’s manual. People should be able to just talk to it and figure things out!” However, I’ve since come to believe that, while AI systems should be highly intuitive, they benefit from a structured onboarding process just like human employees.

Now, our client engagements include not only development of the AI system, but also tutorial videos, training materials, and other collateral to shorten users’ learning curve and ensure they get the absolute most from time spent interacting with the system. This has done a lot to reduce the “blinking cursor syndrome” people get when faced with a blank input field rather than a more guided user experience.

And we extend this philosophy to the overall engagement with the organization, starting with expectation-setting in sales meetings. For example, while pitching an AI copilot for nurses to a large health system, an executive demanded “100% accuracy: no hallucinations, ever!”

Of course, 100% accuracy isn’t something an AI system can achieve, any more than a human consultant can. So we reframed by asking the executive, “How often do your experienced nurse supervisors misstate facts or give incorrect guidance on this particular subject?”

Their answer: “About 15–20% of the time. That’s why we want to build an AI copilot.”

“So when one of your senior nurses gives an incorrect answer, do you fire them and throw them out of the hospital?”

The executive’s face lit up, and eventually we agreed that a system delivering 99.9% accuracy, while providing training for users and appropriate disclaimers in the agent’s workflow, would still be hugely valuable.

That exchange illustrated something we’ve encountered repeatedly: clients’ satisfaction with an AI system depends greatly on their expectations, and sometimes those expectations need to be managed.

Going back to my own agents, sometimes Weston has given great advice that helped me navigate difficult situations, but other times I shake my head at his output, close the window, and follow my own best judgment.

This isn’t a knock on AI: every human boss, colleague, business coach, and hired consultant I’ve worked with has given both helpful and unhelpful advice. But that’s all right, because I don’t expect my mentors to be infallible, and I don’t expect my AI advisors to be infallible either. I value them because they help me see things from other perspectives and, in doing so, make me better at what I do, to the point where I couldn’t imagine working without them.

5. The Race Will Be Won On the Track… and in the Garage

Whenever I talk about building specialized AI systems, someone in the room will point out that everyone’s using the same handful of AI models underneath.

And that’s true: the AI systems we build leverage a combination of OpenAI, Anthropic, Google, DeepSeek, Moonshot, TypeSafe, SpaceXAI, and Zhipu models (yes, some of the same labs that went after Claude… we did say it was knives-out out there). But saying “AI systems all use the same handful of models” is like saying the Monaco Grand Prix is a race between fifty engines.

The engine matters, obviously. But so do the chassis, aerodynamics, suspension, race strategy, and the team’s ability to make everything work together.

Put another way, Isaac the AI agent is the driver, out in the spotlight working with clients. But behind Isaac’s client-facing interface is the consulting equivalent of a racing team. There are domain experts developing and challenging the methodologies he follows; technical experts improving his software architecture; researchers, analysts, and content managers curating the information he depends on; and evaluation systems making sure his advice remains professionally sound. Some of those contributors may be other specialized AI systems, working alongside humans to develop and test new approaches.

And just as in Formula One, winning this year’s championship doesn’t entitle you to win next year’s. Every competitor is trying to build a better machine, recruit better talent, and find an advantage the others haven’t spotted.

I fully expect something similar to happen in expert AI.

While some people argue that the big tech companies will produce universally competent AI systems capable of customizing and optimizing themselves for each and every organization, that would be a departure from how consulting (and software) has worked in the century-plus since Arthur D. Little, Edwin Booz, and James O. McKinsey founded the first modern consulting firms.

Just as no one firm has dominated any consulting niche forever, we won’t necessarily have a single dominant practitioner in every domain. We may have competing schools of professional intelligence, with different philosophies, methodologies, reputations, and strengths. One consulting firm’s AI might favor aggressive operational improvements, another might emphasize reliability and risk reduction, and a third might specialize in complex situations where the conventional wisdom breaks down.

Of course, there won’t be room for every firm in the winner’s circle. Some firms will establish (or re-establish) themselves as the trusted brands in the new AI-powered paradigm. And the economics of AI will allow those firms to serve more clients, meaning fewer opportunities for firms that can’t offer a substantive alternative (or can’t market their alternative as effectively). There will also always be the possibility of clients bringing certain expertise in-house and building their own AI practitioners, though not every client will opt for that when firms offer vetted turnkey solutions that they maintain and improve on clients’ behalf.

In other words, it’s the same dynamic that already exists with human consulting and traditional software. The only things that will change are the stakes and the speed of competition among players.

And whoever wins any given season, clients are likely to come out ahead. When competing schools of professional intelligence fight for business, the benefits flow to the people they serve: better advice, available at more moments, at a fraction of the cost of a traditional engagement, including for frontline workers and smaller organizations that could never have afforded a consultant’s day rate in the first place.

Conclusion: The Future’s Already Here

None of these predictions requires a technological miracle. Like Alvin Toffler’s books, most of these scenarios aren’t predictions so much as observations about the emerging present.

Forward-thinking firms have already developed AI products. The sales meetings are happening, and contracts are being negotiated. The first AI practitioners are already working alongside clients at their desks. What remains is the much harder work of creating increasingly capable Isaacs that clients can trust, and continuously improving them quarter after quarter, year after year, in order to justify their fees.

Because while not all of the predictions above will come true in every detail, some version of that future has already arrived. The question is how expert consulting firms will reposition themselves to fit into it.