Here in 2026, hype about AI is everywhere. But unlike most hype cycles, a lot of it is true. The problem is it's all so overwhelming. There are two takes we guard against: "AI" is either magic or useless hype. We focus on: how do you actually get value from AI for my real-world business?
In consulting, one of the keys to a successful engagement is to avoid over-promising. Here are ten rules we use as we discuss solutions and deliverables with clients. We stay grounded in what is deliverable while acknowledging that AI is truly a new technology that helps with everything requiring intelligence or understanding.
The key tension: the technology is genuinely powerful, but it's powerful in specific ways that aren't obvious until you've worked with it. These rules are the distillation of that experience.
The Rules
Rule 1: You're Always Solving a Specific Problem
You can't say "solve my business." You can say "I have these five data sources and I need them merged into actionable knowledge." An AI system can do that.
This is the most common misunderstanding. People expect general intelligence and get disappointed. But when you narrow the aperture to a specific, describable problem, the results are often stunning. The discipline is in the scoping, not the technology.
Rule 2: Intelligence on the Outside, Process and Code on the Inside
The best use of intelligent systems today is working out a coded process. Software is a promise — it's an algorithm, it does things exactly one way. When you use an intelligent system to write software rather than just chatting with an LLM, you eliminate variability. You get a well-known function that runs mechanically, every time.
This is the design pattern: use the intelligence to figure out the solution, then crystallize it into code or process that runs without intelligence. Intelligence is expensive and variable. Code is cheap and deterministic. Use each where it's strong.
Rule 3: AI is Very Good at Categorizing
Concrete example: you have hundreds of contracts. The system can find all terms that appear across all of them. A human reviews and says "these are fine, we accept these." Then you find terms that don't match — here's alternative language. Then find things that appear infrequently — oddities scattered across your contracts that you may not even know about but probably want a position on.
The pattern is always the same: intelligence at the edge to categorize and surface, then push it into a process where humans apply judgment. The machine does the reading. Humans do the deciding.
Rule 4: 80% of the Work in 1/100th the Time
This is the compounding rule and it's the most important one to internalize.
Most clients start out not having a sense of how their problems map to what AI can do. You can't just tell AI, "Do what I mean!" But you can tell it what processes take too long and work with it on correlating, categorizing, and condensing raw data into information a human can use. The 80-20 rule applies here. The goal is to get the 80% down to minutes instead of hours, days, or weeks.
How it compounds — the playbook: take five processes that take too long and consolidate 80% of work to minutes — you've changed the rate at which your business operates. People still do the last 20%. But you keep chipping away. And the velocity compounds.
Historical parallel: this is identical to the "software is eating the world" phase. Insurance companies that digitized their paperwork in the 50s-70s could scale. The ones that didn't, couldn't. Same inflection, different technology.
Rule 5: Start With Problems You Are Willing to Solve
This is the secret of a successful engagement — start with the problems your organization is willing to solve. Look at the boring parts. The ugly parts. The parts of your business you don't want to examine because they're messy. AI has no ego, it doesn't know the difference between janitorial and board-level work.
There is enormous gain hiding in the processes nobody wants to touch. Apply the earlier rules: intelligence on the outside, code on the inside, cut the 80%, make humans radically more efficient on the remaining 20%.
Rule 6: AIs are Really Good at Context
If you put your employee handbook, travel rules, vendor rules, and supplier information into one AI-accessible system, any staff member can get context on any of it instantly. The system holds more context than any single person can, and it's available on demand.
This is the "institutional knowledge" problem that every growing company hits — and it's one of the places where intelligent systems provide immediate, obvious value with almost no process change required.
Rule 7: They're Very Good at Rationally Comparing Narratives
Concrete example: a three-party negotiation conducted over email. Two parties were simply talking past each other. Rather than having to mediate between the parties, we gave the thread to an AI and asked it to "make a table of what each party is saying and where they're not acknowledging the other's concerns." Seeing the points of agreement as well where they're blind spots resulted in immediate clarity on where the conversation was stuck.
The parties both wanted a deal and resolved the gaps without any drama.
The AI sees the whole conversation at once and can surface the structural disconnects that humans miss because they're inside their own point of view.
Rule 8: AI Expands the Value of Tools You Have
Tools you pay for. Most software has many times more features than are ever used. Using AI, organizations have a major opportunity to take better advantage of the tools they already pay for. Efficiency gained this way is largely "free."
Not only can an AI read all the manuals, examples, and user guides, it can also actively map the software. The AI can loop with users on common tasks and both suggest better workflow and in many cases, perform the workflows. AI can write "programs" in Excel, specialized scripts for software you already use or even talk directly to the APIs most systems have, but few users use.
Two sub-patterns here:
Complex configuration. AI can walk you through filling compliance PDFs, configuring complex tools like AWS or GitHub — even when the jargon is thick and option-count is huge. This is the difference between getting something done in 30 minutes and never attempting it because you don't have the context.
Specialist tools nobody knows about. FFmpeg, Ghostscript, document generators, command-line PDF tools — these are specialist tools that a tiny percentage of computer users even know exist. The AI knows them all. You can broaden your awareness of what tools exist, and the AI can operate them for you.
Rule 9: AI is Good at Articulating What You've Got
This is "if you can't measure it, you can't move it" made practical. With AI it's possible to get everyone's input, both the leadership's "how it should work" and line-staff's "how it does work."
You don't have to get it all correct at once. Over many sessions, you can describe your process flow to build up the complete picture. People know different parts of the process, remember details they forgot about in the meeting, and ignore legacy workarounds. An AI helps you build the complete picture incrementally and is never frustrated by an ever-evolving process.
Once you have the full picture:
- You find the levers to move your processes
- You understand your real relationship to your clients and vendors
- You can write and update process documentation in "found" time vs. big meetings that quickly become fatiguing.
After a few years of operating, most businesses never attempt to gain a deep understanding of their evolved processes. It's enough that the business is running and the people in it know how to do it. With AI, you can pull all of the pieces together without the expense and disruption of bringing everyone into a room for a day to map current process flows.
Rule 10: Identify and Remove Time, Toil and Tears
What you do every day is more impactful than what you do once in a while.
Business models are crazy about Key Performance Indicators — KPIs. To find things you can improve with AI you only need three KPIs: Time, Toil, and Tears.
Among the great features of using this model is you get to leverage complaints, which everyone has. The problem with complaint is it's fatiguing and many things come up that are not really solvable. But when you have AI mediate these things, it's much easier to focus on what is possible.
The Role of Consultants
Businesses don't have spare capacity. There's no team sitting around waiting for a new initiative. Everyone is busy. Everyone has their process. People organize specifically to not get overwhelmed by the million demands of the business.
The consultant's role: come alongside people who have work to do and can't take two weeks off to pioneer a new process. In short, focused engagements — figure out what would be better, how to do it in a way that eliminates time, toil, and tears, how to test it so you know it's actually better, and what the new standard looks like.
If you could just turn on an LLM and ask it what to do, you wouldn't need consulting. We're not at that point of the revolution. We're at the point where we take 80% of repeatable work and radically cut the time and radically increase the quality. That's the place to start.