I Built Two AI Strategists to Argue About My Company

Last week I wrote about the permission problem: the structural tendency of AI systems, and of most feedback environments, to default to accommodation rather than challenge. I ended with a question: have you designed your environment to produce honest feedback, or just to produce feedback that feels good?

This week I want to tell you what I did about it.

The short version: I built two AI strategists, gave them opposing mandates, and asked them to evaluate the current positioning of my company, Move Forward. One was instructed to build the strongest possible case for everything that was working. The other was instructed to identify, without softening, everything that wasn’t.

Then I asked them to agree on a set of concrete proposals. And a third synthesized the output.

Here’s what happened.

The design came directly from the insight in that mentor conversation. If the default is accommodation, you need to make disagreement structural, not dependent on permission granted in the moment, but built into the architecture of the exercise.

A single AI asked to “evaluate my company honestly” will produce a balanced assessment that leans positive. This is not a bug. It’s how the system is trained. The way around it isn’t to ask harder or to phrase the question more aggressively. The way around it is to separate the function: give one agent the explicit job of finding what’s good, give another the explicit job of finding what’s broken, and don’t let them merge until they’ve each done their work.

The structure was:

Agent 1: The Advocate. Mandate: identify every genuine strength in Move Forward’s current positioning. Market fit, differentiation, proof points, momentum. Build the best honest case you can for why this is working.

Agent 2: The Critic. Mandate: identify every vulnerability, gap, and problem in Move Forward’s current positioning. Assume a skeptical investor or competitor is reading this. Do not soften.

Round 2: The Negotiation. Both agents review each other’s output and agree on a prioritized list of changes. Where they disagree, they must state why.

Agent 3: The Synthesist. Takes the full conversation and produces an actionable summary: what to protect, what to fix, and in what order.

The Advocate’s output was useful in a way I hadn’t anticipated. I expected it to produce a list of things I already knew were working, a flattering inventory. Instead, the forced positivity of the mandate made it identify strengths I had been underselling, or not selling at all.

A few things that landed:

The positioning at the intersection of technology and leadership is genuinely rare. Most consultants either understand the technology or understand the human side. The combination is harder to find than I’d been communicating, and the Advocate was clear that I was treating it as a given rather than a differentiator.

The accumulated experience, nearly two decades in digital, across multiple sectors and geographies, was being presented as background rather than as the core of the value proposition. The Advocate flagged this as an active positioning error.

The existing client relationships had a depth and durability that wasn’t reflected anywhere in how Move Forward described itself externally.

These weren’t surprises, exactly. But seeing them listed, argued for, and treated as genuinely valuable was different from knowing them abstractly. The Advocate’s job was to make the case, and making the case required being specific in a way I hadn’t been.

This is the part I’d been avoiding.

The Critic’s mandate was to find what wasn’t working, and it did. A few things I can share:

The external positioning of Move Forward was described as “too broad to be memorable and too narrow to be obviously scalable.” The intersection of technology and leadership is genuinely a positioning, but without a sharper angle, it reads as a description of a person rather than a value proposition for a client.

The content output, including this newsletter, was identified as high quality but inconsistently connected to the service offering. What does reading Beyond The Code lead someone to do, specifically? The Critic’s answer was: nothing obvious. The content exists in a separate track from the business.

The go-to-market approach relied too heavily on network and reputation, which works until it doesn’t, and which doesn’t scale in the way the business ambitions would require. The Critic was direct: this is a ceiling, not a strategy.

Reading this was not comfortable. It was also not surprising, which is its own kind of data. The Critic wasn’t telling me things I didn’t know. It was telling me things I had been avoiding making explicit.

The negotiation phase was where the exercise became most interesting. Both agents had access to each other’s outputs. Their task was to agree on what mattered most.

The agreement was faster than I expected, and it landed on three things:

The positioning needs a sharper anchor. Not a tagline. A clearer answer to the question “who is this for and what specific problem does it solve?” The intersection of technology and leadership is the territory. It needs a more specific address within that territory.

The content and the business need to be connected. Every piece of content should have a legible line to what Move Forward does for clients. Not promotional, structural. The reader should be able to follow the thread from the idea to the application.

The network dependency needs a parallel track. Not abandoned, network and reputation are genuine assets. But supplemented with something that doesn’t require a personal introduction to initiate.

These are real things to do. They are also things I had known, in a general way, for some time. The value of the exercise wasn’t in discovering them. It was in having them stated clearly, argued for by two opposing perspectives, and agreed on: which made it significantly harder to continue not acting on them.

I want to be honest about what this kind of structured AI evaluation is and isn’t.

It isn’t a substitute for external perspective. The AI can only work with what you give it: your description of the business, your framing of the problem, your selection of what’s relevant. It has no access to your market, your clients, or the competitive landscape beyond what you’ve described. The Critic can only criticize what it’s been shown.

What it is, is a reliable way to make explicit what you already know but haven’t confronted directly. The value isn’t in the information, it’s in the structure that makes the information unavoidable.

The Advocate makes you articulate your strengths with specificity. The Critic makes you name your vulnerabilities without the softening that usually accompanies them in your own internal monologue. The negotiation forces a prioritization you might otherwise defer indefinitely. The synthesis gives you something to act on.

And the whole thing takes, depending on the complexity of what you’re evaluating, somewhere between one and three hours. That’s a cheap price for a conversation you’ve probably been avoiding for months.

Last week I wrote about a mirror that only shows you what you want to see. This week is about building a different kind of mirror: one with two angles, one flattering and one not, that together produce something closer to an accurate image.

The AI didn’t tell me anything I didn’t know. But it said it in a way I couldn’t dismiss, defer, or absorb back into the comfortable background noise of knowing-but-not-acting.

That’s what the design was for. And that’s why it worked.

If you want to run a version of this yourself, here is the structure I used. Adapt the context to your situation.

Agent 1: The Advocate

“You are a senior strategist whose sole job in this exercise is to build the strongest honest case for [company/project/idea]. Your mandate is not to be balanced — it is to identify every genuine strength, every real differentiator, every proof point that supports the argument that this is working. Be specific. Do not hedge. Context: [describe your business, positioning, and current situation in as much detail as relevant].”

Agent 2: The Critic

“You are a senior strategist whose sole job in this exercise is to identify every vulnerability, gap, and structural problem in [company/project/idea]. Assume you are advising a skeptical investor or a well-informed competitor. Your mandate is not to be balanced — it is to find what isn’t working, what is missing, and what poses a real risk. Do not soften. Context: [same description as above].”

Round 2: The Negotiation (give both agents each other’s output)

“You have now read the Advocate’s case and the Critic’s case. Your task is to agree on a prioritized list of three to five concrete changes. Where you disagree with the other perspective, state why. The output should be specific enough to act on.”

Agent 3: The Synthesist

“You have the full conversation between the Advocate and the Critic, including their negotiated proposals. Produce a clear summary: what should be protected, what should be changed, and in what order. No more than one page.”

One note: the quality of the output depends entirely on the quality of the context you provide. The more specific and honest you are in describing the situation, including the things you’re uncertain about, the more useful the exercise becomes. Vague input produces vague output, regardless of how well the structure works.

What’s the thing in your business, or your leadership, that you know needs addressing but haven’t made explicit yet? Sometimes the hardest part is just naming it.

Leave a comment

Reply here or find me on LinkedIn.