Tomi Virtanen, Author at Rework

Process mapping has long been a reliable tool for improving processes. You observe the work, interview the people involved, and document the steps, systems, handovers and bottlenecks. You find where time is lost, where quality suffers and where the experience could be better.

That still matters. But in AI-native redesign, it is not always enough, because process work starts from the current process. It assumes the work will remain recognisable, and aims to make it faster, clearer or more consistent. That is valuable when the goal is improvement, but not when the goal is redesign.

So instead of asking only “How does this process work today?”, we also need to ask “How would we achieve the desired outcome if we started today?” That is a harder question for everyone involved.

It is hard for the people doing the work, because they know the current reality deeply: the exceptions, the informal fixes, the difficult customers, the missing data, the unhelpful system screens and the small workarounds that keep the process alive. That knowledge is essential, but it can also make radical redesign difficult. When asked how their work should change, people often answer from inside the current constraints. They may not know what AI can already do, or which parts of the work could be moved, removed or reshaped. They may also have good reason to be cautious. If the discussion sounds like a cost-cutting exercise, they will defend the current process rather than imagine a new one.

It is hard for consultants too. Traditional methods reward careful listening, structured analysis and incremental recommendations: map the current state, find the pain points, propose improvements. Often the organisation already knows what needs to be done, and the consultant is there to confirm it is on the right path and package the ideas into a coherent plan. That can still produce value, but it may miss the bigger shift. In AI-native redesign, the consultant cannot only ask what should change. They also need to bring new possibilities into the room, not as technology hype or a list of tools, but as concrete choices about how work, interaction, engagement and transaction could operate differently.

AI changes more than the sequence of activities inside a process. It can change the whole system of interaction around the work. In many traditional processes, people search for information, move it between systems, prepare decisions and complete transactions step by step, interacting with the organisation through forms, emails, portals, meetings, tickets and handovers. AI makes it possible to rethink that model. The interaction can become more conversational, contextual and continuous. The system can ask for missing information, assemble context, prepare the next best action, surface evidence, assess confidence and trigger the right workflow before a person has to start from a blank page.

The real design question is not how to make the process faster. It is how work is distributed when people, AI and systems share it. Does the employee fill in a form, or does the system carry them to the outcome? Does the expert produce every answer, or stand over the ones AI produces? Is the work standardised for control, or personalised because AI makes personalisation scalable? Do people start every transaction, or does the system start the ones it can see coming?

Each question reframes a default as a choice. Together they describe a shift from designing steps to designing judgement.

Current-state mapping alone will not answer these. They require imagination, but not fantasy. The work still needs controls, governance, trust and clear accountability, and the new model still has to be economically viable. In many cases the answer is not full automation but a better division of labour between people, AI and systems. To scale the benefits, it often means rethinking the underlying architecture as well and upfront investments for wider portfolio of use cases.

That is why AI-native process work should not be reduced to automation opportunity mapping. An automation opportunity is usually a task. Radical redesign is about the shape of the work. A company can automate individual steps and still keep the same slow, fragmented process. It can remove manual effort and still leave decision-making unclear. It can introduce AI tools and still fail to change roles, measures, accountability or the way people actually engage with the organisation.

Leadership should therefore be careful about asking only “Where can we use AI?” A better question is: “What work should still exist in this form? How should interaction, engagement and transaction change? Where are human judgement and accountability non-negotiable?” That is a more demanding conversation, and it is where the strategic value of AI becomes visible.

At Rework, we help companies look at AI opportunities on three levels, optimisation, radical optimisation and redesign, depending on their objectives. Sometimes the right answer is to make the current process faster or lighter. Sometimes it is to rethink a larger part of the flow. And sometimes the real value comes from redesigning the work itself: the roles, decision rights, interaction model, service experience and transaction logic around it.

We will soon share examples from client work where AI has shifted the conversation from automation to operating model redesign. If you want to understand which level your organisation should be aiming for, we would be happy to talk.

AI Knows Everything From Nowhere

In the last post about why imitation isn’t quite enough in consulting, we established, almost as a passing remark, that to pass the Turing test now, an AI would need to be made less capable, not more. Maybe that idea is worth dwelling on a bit longer. Made less capable. That is, more limited in what it can do? Why would limitation be what makes a system pass as a person? And what would that tell us about what a person is?

Let’s say you start exchanging messages with someone who is either a state-of-the-art language model or a random human. How do you know who you are in correspondence with? I know what my first instinct would be: I’d ask it trivia questions. Not to see what it knew, but whether it knew too much. The capital of Assyria. Got that one? Fine. Then I’d go to Wikipedia for harder ones. What is the atomic number of tungsten? The year Krakatoa erupted? The currency of Bhutan? Oh, you knew those too?

The giveaway is that the LLM is not positioned. It has no biography, no preference. Every available piece of information is equally available. In this regard, it is like an automatic typewriter that takes whatever has been written so far, and proceeds with the most likely continuation. The text has no author, but more often than not, it ends up containing the right answer.

Human cognition, on the other hand, is positioned. Very concretely so.

Think of the room you’re in. Maybe there’s a table, a cup, a chair, all within your reach. Outside, a bright morning, or an evening drawing in. A solid floor under your feet. All of this is present and available to you in a way an item behind the wall or on some narrow alley in the neighboring town isn’t.

To you, these things are here, while most things aren’t. Why is that? Had you walked somewhere else, into some other room, you’d have something else at hand. The cup and the chair would not be a part of your world right now.

From your perspective, I’d say it is your biography up until this point that decides what is there.

For an AI to pass the trivia test, it would need something similar. Leave out experience, keep the structure: a biography that decides what is available to it right now. Not a training run, from which the model emerges with full, equal access to every data point there is, but a curriculum, a history it has actually moved through. Something is happening to it now, and the rest it can recall from its past. And then, of course, there are the things it has never encountered. Who was the Doge of Venice in 1655? No idea, I have spent the last cycles reading a software engineering book. But I can google it, of course.

This is not about constructing an inferior AI, but about it having a position. To know something is to know it from somewhere, and from somewhere most things are out of view. A point of view, if we want AI to have one, lives in that contrast.

Trivia answering capability aside, there will probably be markets for more than one kind of system. A great deal of what we need is best done with no point of view at all. Translate this contract, pull the dates out of it, turn this specification into working code, summarize these thousand reviews. Here you want the same input to yield the same output, uncolored by whatever the system happened to read last. The lack of a position is the whole point. You want the right answer and nothing else, and the automatic typewriter, more often than not, gives you exactly that.

ChatGPT arrived in 2022, and the models and tools that followed have grown steadily more fluent and productive. But one thing has not changed: the models themselves are as stateless as ever. They remember nothing from one exchange to the next. Whatever of a weeks-long correspondence still fits in the context window is handed back to the model on every turn, and it reads it fresh, as if meeting it for the first time. And for the work we just described, that is exactly right.

There is the other kind of work, though. The kind we are building Rethink for. Think of a lawyer halfway through a case. Her work is not to know everything. It is to know this case better than anyone in the room. To hold what she has seen of it, to weigh this document against that one, to notice when something does not fit the story so far. She does not read a transcript of the entire engagement every morning. Her experience is biographical. This is the kind of work where a point of view stops being a curiosity and becomes the job.

Consulting Needs Subjective AI

If I had to name a single thing that has been most surprising in the rise of AI in the last few years, it would be how badly the Turing test ended up aging.

For a long time, it was treated as the cleanest imaginable threshold between the mechanical things computers do and real human intelligence. If a machine could converse so naturally that we could not tell whether it was human, surely something important had been reached. And surely, at some point, there would have been big headlines about the test having finally been passed.

But I don’t remember ever seeing those headlines, and now it’s probably too late for them. We seem to have moved past the point almost without noticing. And more often than not, the AI would now have to be made less capable, less articulate or at least a much lazier typist, to pass for a human.

That is probably what happens when the AI breakthrough arrives through language models. They are built from patterns in human expression, so the first thing they did was to become remarkably good at reproducing the surface of human conversation. The Turing test was looking for imitation, and suddenly imitation became the easy part.

Sounding Like a Consultant Is Not the Same as Being One

This matters for AI use in fields where the answers are supposed to carry real consequences. Management consulting, for one, is almost defined by high-stakes ambiguity: important decisions, incomplete information, conflicting evidence, and people who will have to live with the outcome.

In that setting, imitation is not enough. A system that merely sounds like a consultant is a little like three children in a long coat trying to pass as an adult. From a distance, the silhouette may be convincing. The suit is there. The posture is there. Maybe the vocabulary is there too: operating model, value levers, strategic priorities, decision velocity.

But if you were deciding how to reorganize a business unit, where to cut cost, or whether to bet months of work on a new operating model, the suit and the vocabulary would not get very far. You would want to know whether there is actually a responsible point of view underneath it. Whether someone with the experience and capability to grasp the problem has actually done the thinking.

This is where LLM-generated consulting output falls short as advice. It can reproduce the surface of the profession: the tone, the vocabulary, the structure of a memo, but the style is only useful when there is a grounded read underneath it.

An Artificial Consultant Needs a Point of View

Put it another way: would you be happy paying a consultant who quietly pasted your problem into ChatGPT, copied the answer into a slide deck, and presented it as their own judgment?

Probably not. Not because ChatGPT is useless, but because that is not what you thought you were buying. You were paying for someone to understand the situation, weigh the evidence, take responsibility for a read, and tell you what they can actually stand behind.

That missing layer is what I mean by a point of view. The consultant is not paid merely to produce the final words on the slide. They are paid to stand somewhere in relation to the problem: to know what they have seen, what they believe follows from it, what remains uncertain, and what they would refuse to claim. If AI is to do more than draft the slide, the same standard applies.

In practice, that means the system does not treat every question as an invitation to produce an answer. It must first ask what the engagement so far actually supports.

Ask it for the payback on an initiative. If it only has cost-side observations, but no benefit measure, no time horizon, it should not invent a fluent payback story. Cost data does not become payback data just because the question was asked in payback terms. A system with a point of view says: I cannot stand behind that claim yet. The strongest adjacent claim I can support is the cost denominator, and here is what would close the gap.

Or take disagreement. Suppose the interviews say the review process is blocked because one team is overloaded, but the workflow data shows delays scattered across several handoffs. A fluency-first system may blend those into a vague summary: “capacity and coordination issues are slowing delivery.” A system with a point of view should keep the signals separate. It should say: the interview evidence points to team overload; the process evidence points to handoff fragmentation; these are not the same diagnosis, and I would not collapse them into one recommendation yet.

Or take decisions. If leadership accepts a recommendation, that decision should matter in future work. But it should not magically prove that the original diagnosis was correct. The decision becomes a constraint, while the evidence trail stays open.

That is the difference. A fluent system tries to answer the prompt. A consulting system with a point of view knows what it can stand behind, what it cannot yet claim, and what has changed because of earlier work.

This Is the Standard We Are Building Rethink Toward

At Rework, we are building Rethink to support the kind of consulting work where fluent answers are not enough. The goal is not to create another system that can produce polished business language on demand. The goal is to build an artificial consultant that can participate in the work with a grounded, inspectable point of view.

If Rethink says a claim is supported, the evidence should be inspectable. If it refuses a recommendation, the missing input should be clear. If its read changes, the reason should be traceable. If leadership accepts a decision, Rethink should remember that as part of the engagement without pretending the evidence question is closed.

That is the kind of system we would be comfortable attributing work to: one whose work can be challenged, corrected, and carried forward without pretending the system is human.

This is why we have started using the phrase subjective AI at Rework. The wording is a little risky, because subjective often means biased or merely personal. That is not what we are after.

We mean something more specific: AI that has a position inside the work. It has seen some things, missed others, made some commitments, and changed its read over time. Its judgment is not floating above the engagement. It is part of the engagement.

In fact, one of the ways I will know Rethink is working is that it will not always answer me straight away. Sometimes it should pause. Sometimes it should ask what the engagement actually supports. Sometimes it should decline to draw the process diagram I asked for, because the evidence is not there yet.

That is what we are building Rethink toward.