I want my competitors to all use AI in its current state in their strategy. Slather it everywhere, make all your product strategy auto-generated from auto-polled sources like auto-summarized customer feedback.

I’d rather they be wrong without wasting that energy and the other ethical concerns, sure. But I’ll take them being wrong to start.

I tried strategy as part of my ongoing “try using AI tools for product management” experiments, and LLMs today tell everyone to do the same thing, and worse, that thing is to drive your car right into the nearest wall, and then the next-nearest larger wall.

“Blue Ocean Strategy” is the most useful way I can think of to explain why their advice is terrible. Blue Ocean Strategy says “pursue differentiation”: find products and services you can offer that your competitors can’t or won’t. This is contrasted against “Red Ocean” markets, where everyone’s viciously competing and tearing each other apart, spending much to little or no gain.

In good product management, you’re considering what your strengths are, what you have on hand, what you can do with it, how much time you have, what you know, all that good stuff, and then figuring out what customer problems you can solve in a way that help them and make for a sustainable business. If you do that, the opportunities to do something different, do it better than your competition can, while you make customers happy and the business successful… the ideas will come to you with work.

The opposite of that is to try and build an exact copy of a huge successful existing business. I work in fintech, and I frequently meet people who want to start the next Venmo or Paypal. I try to ask “can you find a niche that you can serve much better than them, even if it’s much smaller, to start with?” I used to try and explain how hard it is to build those network effects, how well-funded they are, how sharp they can be at defending their turf, and how slim their chances are.

LLMs tell me, and my competitors, that that’s a great idea.

My current company is good at orchestration and business logic for payments, adding new partners and services, and building good APIs. We don’t have full banking-as-a-service or card offerings, for instance, and we’re tiny, our engineering capacity is very low. Our biggest product today is a true “call us once to make a bank payment and we’ll do all the lifting to make it get there, including failing over to other methods” where we take the headaches and complexity off our customer (and give them to me!).

I fed some prompts into different LLM tools, and they all told me I should ram the company into the largest, nearest competitor, often followed by ramming the company into the second-largest competitor afterwards. Which if you think of LLMs as most-probable-next-word prediction enginges gone nuts, makes sense: those are the use cases with the most money and adoption behind them, they get written about the most.

Ignoring the frequent suggestions to do things we already do, they told me:

• Embedded finance
• Become a vertical provider
• Get into stablecoins and digital assets. Tokenize all things!
• AI-driven financial services (automation/payments/compliance/etc)
• build a superapp!

There were also themes in potential customers they mentioned – they all wanted to get into gig economy payments, for instance. This would come up as “move into vertical SaaS” (or vertical __)”

I started with “based on what you know about my company… what should we try next for product strategy?” then for the first time in this series, I tried different prompts, including adding more information saying “we’re good at these things, we have previous experience here” and getting into much more detailed background and trying to set them up to give more nuanced answers.

Let’s talk about getting into “embedded finance” for instance. Stripe has, in my opinion, the best zero-to-taking-money experience in the industry, from introduction docs to the path you take as a new customer in signup. They’re great, and it’s built on years on years of often unglamorous, unseen work by smart people grinding it out. There are things they don’t do as well, or have chosen not to go after, that you could go after.

And that’s one of the established, competent players who already have market traction.

Similarly, stablecoins are hot again, and the players in the space are newer, but whatever you’re looking for, there are a lot of more-established, smarter players in those spaces. Circle, for instance, has about 200 people and a US-dollar stablecoin with wide adoption ($60+ billion in circulation as I write this).

The superapp thing… ChatGPT 4.1 said

Unified Financial Operating System: Build a “superapp” or platform that consolidates payments, digital wallets, lending, compliance, and analytics into a single, customizable environment for fintechs and non-fintechs alike

And Gemini 2.5 Pro said

This strategy capitalizes on the global superapp trend, which is reshaping consumer expectations. Many companies want to offer an integrated experience but lack the resources to build one from scratch. Sila could provide the foundation.

Where do you even start?

(Super-app answers seemed to be inspired by the same post every time it came up, which led me to a “how are these being drawn? Is this something with Perplexity?”)(is it? Please let me know)

I did at this point think some of the “go build something you have” advice, and the ill-suited is because the information on our tiny startup is pretty bad, and many of the sources the models are seemingly drawing on are out-of-date, or confused (there’s a much larger battery company also named Sila, for instance, so some sources think we have 40 patents around that, or that we raised $375 million in a Series G round). That’s a separate issue, and “how do I update the big LLMs on my business” is something almost everyone’s trying to figure out.

I stopped here: the models seemed to want my company to do what everyone else was doing, or was about to start doing, and thus compete with established players who were already there and would crush us.

There was a little value in some of the answers sparking some deeper thought for me — I’m going to think about how they came up, I’ll report back on whether this turns out to be a way to generate threads to pull on.

Before I threw my hands up, though, I wanted to see what strategic advice other players in the space would get given the same high-level prompts, and it was… the same.

Banking as a service company with a banking charter (excerpt from Claude)

• “AI-First Banking Infrastructure”
• “Target verticals underserved by generic banking APIs (e.g., healthcare payments, real estate, gig economy, B2B SaaS).”
• “Full-Service Platform: Launch a “bank-in-a-box” for startups—bundling compliance, KYC/AML, payments, lending, and analytics.”

(and here I’d say “they do that last one already”)

Adjacent not-direct-competitor, not a bank (excerpts from Gemni answer)

• “Vertical-specific Specialization” (insurance, gig economy, brokerage and wealth management, Vertical SaaS)
• Focus on developer experience
• A super-app pitch (“Shift focus from selling the API to selling a complete SaaS solution that is powered by the _ API.” )
• The “Embedded Finance” Enabler

Different not-direct-competitor (excerpts from ChatGPT)

• “Deepen focus on Embedded Finance”
• “Leverage AI and data-driven services”
• “Expand into New Geographies and Segments” (What! A unique one!)
• And then suggested some radical pivots, like “Vertical SaaS + Payments”, “offer compliance, banking, lending, and insurance APIs as a unified platform” and — Blockchain and Stablecoins!

Payments processing company (excerpts from Gemini)…

• “Dominate a vertical market” … “Shift from a horizontal platform serving all business types to becoming the undisputed payment infrastructure leader for a single complex industry. Potential verticals include B2B SaaS, healthcare, insurance, or creator economy platforms.”
• Cited a payments product they’d launched and then burned a lot of words on “make it the leading product” without (to my eyes) any real advice that wouldn’t have occurred to them in thinking about the product and then launching it

Few of them even noted when a company enjoyed a particularly strong position — when I saw an answer say something like “you just raised a ton of money, use that to your competitive advantage by investing in product offerings” I was surprised.

I love this for them! If you’re in an industry where suddenly all your competitors are hyper-focused on the same three vertical markets, spending ever-increasing amounts on customer acquisition and launching similar products into a crowded market (and then copying each other’s incremental improvements) wouldn’t you be absolutely delighted? The rest of the field is yours. Listen to your customers. Experiment with interesting products. Spend time mining data for insights.

Watch them crash their burning bumper cars into each other over and over, waving their last dollar at the attendant to let them have another minute driving.

Now you’re thinking there is some interesting utility here, in that as consensus-regurgitators, the LLMs are giving an intresting insight into what we might say is the default strategy. You in turn would want to view any product strategy that took you in that direction as likely to be into greater competition and dimineshed returns – a signal to look for other options.

And you could then take a minute to think “if I was in my competitor’s position, how would I view this advice? Would any of it be tempting? What would I immediately discard?” as a useful thought experiment that might give you an idea of where they’re being pulled by the gravity of consensus.

For now though: AI strategy for thee please, but not for me.

Some open questions:

• I’ve been using Perplexity for ease of generating these from different models. Is that also introducing issues? The “super-app” thing from different models has me furrowing my brow, but in the past when I’ve also generated the answers directly from the model’s purveyors, they’ve matched reasonably closely to what I’ve gotten with calling that model via Perplexity
• Is there a way to give enough context and instruction that makes this exercise go well? Are there different approaches to the prompt itself, or setting this up, that help?