AI Is Making Me Much Faster at Consulting Work. That Makes the Old Value Logic Awkward.

Open Source Consulting began as a content production engine. What surprised me was how quickly the same pattern became relevant to consulting itself, making the question of AI consulting value impossible to ignore. The broader category argument is now captured in AI-Native Professional Services: redesign the service around what remains scarce when research, synthesis and artifact production get cheaper.

I wanted to test whether AI could help me turn ideas, observations and professional experience into a more disciplined publishing system. It did. Once AI can take unstructured material, recognize patterns and turn it into structured recommendations, the interesting question is no longer whether it can help me produce more. The interesting question is what part of expert work still deserves a premium when the production layer becomes dramatically cheaper.

I had seen a version of this problem before. At a previous employer, I once explored whether the unstructured lessons buried in pursuit retrospectives could become something much more useful than a static archive. If enough teams documented why deals were won or lost, AI could look across those fragments and surface recurring patterns about what actually differentiated the company. Instead of asking people to manually read dozens of retrospectives, the organization could query its own experience and receive a structured recommendation about how to sell itself more effectively.

That idea did not become the system I imagined, but the underlying mechanism stayed with me: expensive professional value often sits inside messy, distributed context. The work is not simply to create a document. It is to recognize what matters, connect evidence that was not originally structured for the question, and convert it into a decision someone can act on.

OSC is now making that mechanism tangible for me every day. I can dump notes, conversations, research, decisions, documents and half-formed hypotheses into an AI-assisted working system. I can leave material deliberately unstructured for longer than I would have tolerated before because I know I can structure it when the question becomes clear. The system can maintain context across workstreams, reconcile sources, draft artifacts and surface contradictions. That is useful for publishing. It is also uncomfortably close to a large part of what professional services have historically charged people to do.

The artifact is becoming cheaper to produce before the decision becomes cheaper to make.

A consulting engagement where every minute with the team mattered

One of the clearest examples comes from an anonymized consulting engagement I worked on recently. I had been asked to help a team in India adopt a more agile way of working so that decision quality improved and collaboration with stakeholders in the United States could become more asynchronous. The technical mechanics mattered, but the engagement was fundamentally human. I needed to understand how the team communicated, where hierarchy constrained decisions, which cultural habits made it harder to challenge assumptions, and how much confidence people had to act without waiting for a meeting on the other side of the world.

Every living second I could spend with that team was valuable. I wanted to be in conversations, observe behavior, challenge patterns, coach people through decisions and build enough trust that the new way of working did not feel like a process imposed by an external consultant. That was the scarce part of the engagement. It depended on presence, context and a relationship strong enough for people to tell me what was really happening rather than what the process diagram said was happening.

At the same time, the engagement required regular sponsor reporting and due diligence with stakeholders in the United States. That was entirely legitimate. The sponsors had funded a fixed-fee engagement and needed evidence that the work was progressing, risks were visible and the intervention was producing something more than good conversations. But the reporting layer was repeatable. I was effectively my own PMO.

Without AI, I would probably have needed a more junior colleague to help me maintain that layer. That would have created a commercial problem immediately. The extra resource had not been budgeted into the fixed fee, so I would either have needed permission to reduce the margin or I would have absorbed the cost. Doing it alone would have been close to impossible without sacrificing time with the team. I remember working through four-hour layovers in Dubai and Abu Dhabi because the operational reporting still had to be done after the client-facing work was finished.

AI changed the balance. I could keep the raw material relatively unstructured and continuously feed notes and observations into the system. That material could then be translated into structured stakeholder reports and due-diligence artifacts. I still had to review the output, correct it and decide what deserved escalation. But I could invest roughly 95% of my own attention in the team rather than in the administrative machinery around the engagement. I managed to do six months worth of progress in eight weeks, with the amount of output a three-person team would generate, by myself, because I was reacting and adapting to every need the team had immediately.

This was not a theoretical productivity exercise. The technology changed where my professional attention could go. It reduced the cost of proving that the engagement was under control while preserving the part of the work for which my presence actually mattered.

AI consulting value changes when the work changes composition

I think this is where some of the discussion about AI productivity in professional services becomes too shallow. We talk about hours saved as if a consulting engagement were a uniform block of labor. It is not. It contains different kinds of cognitive work with very different relationships to value.

Research can often be accelerated. Notes can be transformed into structured knowledge. First-pass synthesis can become extremely fast. Scenario generation, comparison and drafting can happen in parallel. A large amount of formatting, documentation and reporting can move from manual production into supervised production. In my own work, translating ideas into structured documents has become so reliable that doing the first 75% manually now feels strangely wasteful.

This does not mean that the output is automatically good. It means the production cost of reaching a plausible first version has fallen sharply.

BCG’s research across legal, tax, accounting and financial professionals found that specialized generative AI tools produced stronger time savings and required less rework than general-purpose tools. The exact percentages are less important to my argument than the direction: as AI becomes better matched to the work and the context, the human contribution shifts away from producing every intermediate artifact and toward deciding whether the artifact is useful in the first place. BCG’s professional-services research is not evidence that every consulting workflow behaves the same way, but it is consistent with what I see in my own practice.

The shift is already visible in client expectations. Thomson Reuters reports that organization-wide generative AI use in professional services reached 40% in 2026, while only 18% of respondents said their organizations tracked AI return on investment. More importantly for the commercial relationship, two-thirds of corporate respondents wanted their outside firms to use AI. That combination of adoption and weak measurement creates a strange period in which clients increasingly expect efficiency without the industry having fully agreed how that efficiency should translate into value and price.

The first 75% getting easier makes the final 25% more important

My own AI workflow now looks very different from the one I used on that earlier consulting engagement. I do not carefully structure every input before the system sees it. I dump material into the working context and structure it iteratively. Notes become knowledge. Knowledge changes the questions I ask. New questions reveal gaps. Those gaps trigger more research or a different conversation. Several parts of the work happen simultaneously instead of as a neat sequence from research to notes to synthesis to document.

The short iterations are important because they reduce the size of any individual error. I am not asking AI to disappear for three days and return with a strategy. I am continuously seeing what it is doing, challenging it, correcting assumptions and enriching the context. That makes the system more useful over time, but it also creates a non-negotiable requirement: I have to know what is happening.

If I choose not to read the AI-generated work, I am making a conscious decision to remain ignorant about my own initiative.

I think this is one of the mistakes behind the false choice between “read everything” and “approve AI blindly.” The real question is where the human must remain knowledgeable enough to judge the system. If I am accountable for the initiative, I cannot outsource my understanding of it. The AI can compress production. It cannot make my ignorance responsible.

That is also why asking the right question has become more important to me by an almost absurd margin. AI can produce a very good answer to a badly framed question. It can also produce ten plausible answers before I have decided whether any of them matter. As the cost of answering falls, the relative value of choosing the question rises.

This is closely related to what I previously called redesigning work with AI: once cognitive effort becomes cheaper, the opportunity is not merely to accelerate the existing process. It is to reconsider which parts of the process deserve human attention at all. The professional-service equivalent is uncomfortable because some of the work that used to make an engagement look substantial may have been expensive primarily because producing it consumed scarce human time.

AI can overdo the work and make the excess look reasonable

There is another reason expertise does not disappear when production gets easier: AI has almost no instinctive aversion to doing too much.

A human questions every key press because every key press requires some amount of attention. AI will happily consume available compute to produce another artifact, another story, another option or another layer of structure. The result can look impressive precisely because the system is capable of filling gaps with plausible work.

I saw this recently when I asked AI to translate an unstructured Excel sheet into a Jira backlog. I had asked for a minimum of 30 user stories. It created 50. On first inspection, the backlog looked productive. It was organized, detailed and apparently comprehensive.

I immediately saw scope bleeding from one user story into another. The AI had not discovered hidden requirements. It was compensating for missing input by generating structure around the gaps. Because I had spent years maintaining backlogs and documenting requirements, the failure mode was obvious to me. Without that experience, the extra 20 stories could easily have been mistaken for diligence.

This is the inversion I find most interesting. AI can make expertise look easier at exactly the moment expertise is doing more of the filtering. The visible effort disappears. The invisible act of saying “no, this is the wrong structure” becomes more important.

That is why I do not think the future value of consulting can be reduced to prompt quality. Better instructions matter, and my own article-production workflow now needs far fewer corrections partly because I have learned to convey my voice and feed the system better knowledge. But learning to steer the system is not the same as learning what good looks like. The steering improves because experience gives me a stronger model of the destination.

The same principle appears in my experiments with AI agent decision quality. Repeated corrections become valuable when they are turned into durable context and operating rules. The machine can improve because the human can explain why a seemingly plausible answer was not good enough.

Trust changes the commercial obligation

The most successful consulting relationships I have experienced were not successful because I produced the most slides. They were successful because the client trusted me enough to tell me what was actually happening and trusted me to use that information in their interest.

That creates an obligation that is easy to overlook in the pricing debate. If I can use technology to reduce cost, increase value and mitigate risk, I believe it is my responsibility to do so. Hiding AI efficiency inside the old delivery model simply because the client cannot see the effort would be a strange definition of trusted adviser.

Professional services have always contained information asymmetry. The provider usually knows more than the client about how the work is produced. That asymmetry is part of why trust matters. AI increases the temptation to exploit it because the visible output can remain the same while the internal effort falls sharply.

I think the better answer is to share the gain.

If I can deliver a stronger result with fewer people and less production effort, the invoice should probably be smaller than it would have been under the old model. But the margin can still be larger because my cost to serve has fallen even faster. The client pays less. I earn more efficiently. Both sides come out ahead.

That is not a universal argument for performance pricing, and it does not mean every professional service can be priced directly against an outcome. Attribution is messy. Providers do not control every variable that determines whether a recommendation turns into value. But the economic direction matters. If I can credibly identify a hypothesis that could remove €150.000 in recurring operational cost, the commercial conversation should be about the value and risk of reaching that outcome, not about whether I needed 17 hours or 47 hours to produce the analysis.

Financial Times reporting this week describes clients already pushing parts of consulting and technology work in-house with AI and increasing pressure for fixed-price or performance-oriented commercial models. The significance is not that one pricing model will replace all others. It is that AI is making the relationship between labor input and client value harder to defend as the default logic of the industry. When you charge for outcomes, you have skin in the game, which is great if you’re building a relationship based on trust.

The premium moves from production to judgment

None of this makes production worthless. Research still has to be grounded. Documents still have to exist. Models still have to be checked. Stakeholders still need evidence they can understand. A badly produced artifact can destroy confidence in a good idea.

What changes is the scarcity premium.

When analysis, synthesis and polished output required large amounts of skilled human labor, they naturally commanded a premium. As those activities become faster, some of that premium should move toward the capabilities that remain difficult to scale: framing the problem, understanding context, identifying the consequential evidence, challenging an attractive answer, building trust, changing behavior and accepting accountability for the recommendation.

This connects directly to the Human Premium Enterprise. My argument there was that as ordinary cognitive output becomes cheaper, organizations need to become much clearer about the human capabilities they recognize and reward. Professional services are simply a particularly visible test case because the client is directly paying for expertise.

The uncomfortable possibility is that AI commoditizes some manifestations of expertise before it commoditizes expert judgment. A beautifully structured analysis may become easy to create. Knowing that the analysis is answering the wrong question may remain expensive.

That is also why I do not want OSC to become a machine for producing more consulting artifacts. The more capable the production system becomes, the more disciplined I need to be about what deserves to exist. If AI gives me the ability to generate 50 user stories, 20 scenarios or a 60-page recommendation, that does not create an obligation to deliver all of them. It creates an obligation to decide which ones matter.

A smaller invoice can be evidence of a better consulting model

I used to think about AI leverage primarily as a capacity question: how much more can one person do? I increasingly think the more important professional-services question is what one person should stop charging a client to do manually.

That is a harder question because it reaches into the economics of the firm. Junior leverage, utilization, pyramids, fixed fees and time-based pricing were all shaped by a world in which a lot of cognitive production required human hands. AI does not remove the need for talent or apprenticeship, but it does challenge the assumption that expensive production is the natural evidence of valuable expertise. I will come back to that broader industry structure separately.

For the individual consultant, I think the obligation is simpler.

On the engagement with the team in India, AI helped me spend less of the client’s money on me being my own PMO and more of my finite attention on the thing the client actually needed from me: helping people work differently together.

That feels like the right direction for AI-native professional services. For me, AI consulting value is not about replacing the consultant with a chatbot or hiding automation behind the old invoice. It is about using AI to compress the commodity layer while preserving the human attention the client actually needs.

Use abundant machine-assisted cognition where it reduces repeatable effort. Use human attention where context, trust, judgment and accountability still matter. Then redesign the commercial model so the client benefits from the efficiency instead of paying for a theater of effort.

If we are being honest, that may mean the invoice gets smaller while the margin gets bigger.

I am comfortable with that. In a trust-based relationship, it may be exactly what winning looks like for both sides.

What part of expert work would you still refuse to delegate?