Rob Kaminski — co-founder of Fletch PMM, the consultancy that has run positioning work for roughly 600 B2B startups including Notion, GitLab and Superhuman — does a live teardown with host Angelina on TwoSetAI (1h02). The subject is a real AI agent startup: AgensFlow, Nicole Königstein’s coordination layer that sits above frameworks like LangChain and LlamaIndex and learns which agent and model to use for each task.

The setup: every AI agent company reads the same

  • Dozens of AI agent startups are launching with the same landing-page words, the same claims, and the same category jargon
  • A great product is not enough to win — buyers can’t remember a product they can’t categorize
  • The remedy is not better copy, it’s a decision about who the product is for and why it’s better

Positioning is a choice about where to be memorable

  • “Living in the customer’s mind” is the output; the work is choosing where you can get in there
  • People sort everything into categories and then remember the leaders of those buckets — that’s how memory works
  • New technology fits multiple buckets (the iPhone was a phone or a computer), and the bucket you pick dictates your channels, your competitors, and your differentiation
  • Categories differ in size, growth rate, and realism: “top five CRMs” returns a list; “top five revenue platforms” returns confusion — that’s a signal of an immature category

Strategy 1 — win a category buyers already know

  • Name the category, then explain why you’re better: “steal mind share using the category as the anchor”
  • DuckDuckGo: “unlike Chrome, we don’t track you” — binary differentiation the incumbent cannot copy because tracking is its business model
  • Figma: to people who already used interface design tools, the pitch was simply “it’s Adobe but multiplayer”
  • Upside: the budget line already exists. Downside: you’re purely in a fight, and standing out is hard

Strategy 2 — sell a job to buyers who don’t know the category

  • When the category is emerging, the buyer doesn’t know the vendors, the lingo, or even the job — so drop the category anchor and anchor on the job to be done
  • DocuSign in 2010: not “the best e-signature platform” but “you’re getting documents signed by courier — don’t do that”
  • Calendly: “you’re scheduling by email back-and-forth — don’t do that.” It ran this educational play for years before switching to “the number one scheduling automation tool” once the market matured
  • This is mostly education: you’re creating the mental category, then claiming it. Greenfield founders often think their competitor is the other startup in their YC batch — the buyer has never heard of either

Strategy 3 — create a category (the one founders overrate)

  • Every founder loves this option; it is the hardest to pull off and carries the most risk
  • You must teach the job exists, then teach the tool for the job, then get a new budget line created
  • The founder of Gainsight, who did it successfully, told Rob: despite all the success, “I will never do category creation again” — a 10+ year journey of pain and education before real money arrived
  • You also can’t ride a hype cycle you invented; and if your “new category” is too close to an existing one (DocuSign’s “intelligent agreement management”), you make your core product harder to understand
  • WorkOS and Vanta show the shape: pure job evangelism when no dedicated category or budget existed yet

Categories come with baggage

  • “AI SDR” is picking up negative connotation from replacement-flavored marketing
  • Call yourself an agent management platform and buyers expect Copilot/Agentforce-style building and governance — features you may not have
  • If the label promises capabilities the product lacks, the category costs you the deal

The bets Fletch weighs for AgensFlow

  • 1A — compete in the real, existing category: position against LangChain and friends as the only self-learning agent orchestration framework. Well-funded incumbents, limited startup resources, and a “do I keep control or is it a black box?” objection to answer
  • 1B — same category, different audience: speak to people starting to build with agents who have never bought an orchestration framework, and sell the coordination job educationally
  • 2 — become an add-on: “a LangChain optimization add-on,” or go broader as “an agent orchestration framework add-on” (needs sub-campaigns because not every platform has the same gaps). Add-on positioning can work extremely well — a company selling an Office 365 add-on for financial advisors does close to $100M
  • 2b — sell the cost job instead: “reduce AI token cost by 70% without sacrificing quality” to a cost-conscious, category-unaware audience. The orchestration is the secret sauce, not the headline
  • 3 — articulate the grand vision: positioning broadly around coordinating and optimizing agent teams. Rob’s caution: can you actually be known for that? Early on the answer is usually no — treat it as an experimentation strategy, and don’t stay in it more than six months

Why “we’re more than X” kills recall

  • The founder’s instinct is to list every cool thing the product does and assume a buyer who knows them all will see the value. That is not how markets consume information
  • Merge two arguments (“best orchestration” + “saves cost” + “self-learning”) and you get a “we’re the best” claim nobody remembers — everyone thinks they’re the best
  • Pick the one sentence someone would repeat at a bar. The other value props layer on inside the sales conversation, where there’s dialogue and time
  • Which single argument you pick should be where the market’s pain is sharpest and where your edge is most defensible — if incumbents are likely to build your feature next year, that’s not the wedge

Positioning is segmentation, not words on a homepage

  • Rob’s definition: target market + why you’re better. It’s a segmentation problem that the homepage merely expresses
  • “A company with six ICPs is either extremely successful with the team to chase six segments, or a delusional early-stage startup that doesn’t know who it serves yet”
  • The forbidden move is serving everyone: more segments means a harder-to-buy product and a less believable story — and nobody knows who you are yet anyway

How long it takes

  • Early stage, pre-traction: positioning may shift every 3 to 6 months
  • Companies with traction: hold the strategy and sharpen it roughly every 12 to 18 months
  • Building actual recognition takes 18 to 36 months of repeating one thing; Fletch took about two years to be recognized, and even then had to shape what it was recognized for (they were known for homepages before positioning)
  • Most founders quit before it works, or run a good strategy for ten days and drift back to being everything to everyone

Landing pages aren’t dying — they feed AI search

  • AEO/GEO is still rooted in SEO: what you publish is what AI systems summarize
  • Describe yourself consistently (~80% of the time) as one thing and AI is far likelier to name you when someone asks who leads your space
  • Call yourself six different things across landing pages and social, and the AI has nothing to hook onto

What actually frustrates him

  • Not lack of insight: the founders he works with arrive with plenty, then refuse to choose. “That’s just bad leadership”
  • A founder he worked with put it perfectly: “All six of these suck and it’s not your fault — they all suck because they all have trade-offs”
  • Founders get paid to make the trade-off decisions; the exercise is choose your hard, then stick with it

“If you wanted certainty, you should have been an accountant, not a founder.” — Rob Kaminski

Books he recommends

  • Rework — Jason Fried and DHH (the one he rereads)
  • Crossing the Chasm — Geoffrey Moore (the roots of Fletch’s models)
  • Built to Sell — John Warrillow (for productized-service businesses)