Skip to main content

Documentation Index

Fetch the complete documentation index at: https://growth.dante.id/llms.txt

Use this file to discover all available pages before exploring further.

By Dante Perea, Founder, unifounder.ai. Founder of unifounder.ai. Building agent-native publishing infrastructure for the Business-to-Agent economy. Previously shipping AI products at the intersection of multimodal models, retrieval, and developer tooling.Follow on X · GitHub · growth.dante.id.

B2A-Native Vertical Agents: The Three-Force Convergence Defining the Next Wave of AI Startups

Three forces are colliding in 2026, and most founders are only paying attention to one of them. The result is a market map that looks crowded on the surface — dozens of “AI agents for X” products — but is actually wide open at the intersection where all three forces meet. The forces: agents are now real customers (B2A), vertical specialists consistently beat horizontal platforms, and the scarce resource has shifted from model capability to organizational authority. The synthesis play is building B2A-native vertical agents — products designed from day one to sell to and through AI agents, in a domain narrow enough to win on workflow depth.

What problem does this solve?

Most founders looking at the agentic AI market in 2026 are picking a single force to bet on. They build “an AI agent for the legal industry” without thinking about how an agent buyer will discover them. They build agent-facing infrastructure without picking a specific vertical to dominate. They build vertical depth without an explicit authority model, then hit the scaling wall when enterprise compliance review begins. This playbook is for founders who want to build at the intersection of all three forces — the position with the highest structural leverage in the current cycle, and the one with the fewest crowded incumbents because most teams only see one or two forces at a time. The clearest leading indicator that this intersection is real arrived on April 22, 2026, in a Fortune piece. Visa’s CMO gave it a name on the record:
“The rise of Business-to-AI, or B2AI. AI agents are becoming a new customer segment.” — Frank Cooper III, CMO, Visa (Fortune, April 22, 2026)
When a payments network CMO names a new customer segment in a major publication, it is a market structure call, not a tech trend observation. Visa’s business is moving money between transacting parties. If they are publicly framing AI agents as a customer segment, they have already modeled the revenue. The financial case is similarly direct:
“Deloitte’s 2026 State of AI in the Enterprise reports an average 171% return on agentic AI investments, with US deployments hitting 192% — three times traditional automation ROI.”
171% average ROI is not a number that requires evangelical selling. Buyers who understand it will find you. The challenge in vertical AI is not convincing buyers that agents deliver value — it is convincing them that your product, in their specific vertical, with their specific compliance requirements, can deliver that value safely.

Before you start

This playbook assumes you are seriously considering building a B2A-native vertical agent — not just researching the space. Before you commit, make sure these conditions are true:
  • You have a vertical you actually understand. The repeating advice across every founder source: “The fastest vertical for you is the one where you have the most connections to potential customers and the deepest understanding of the workflow.” If you do not have either, pick a different vertical or partner with someone who does.
  • You are willing to start narrow. The pattern across vertical AI winners is “pick one narrow niche, dominate it, then expand.” If your instinct is to start broad and narrow later, this position is wrong for you.
  • You can sell on trust, not just capability. B2A commerce treats trust as a ranking signal. Cooper: “Trust will increasingly function as a ranking signal in AI-mediated commerce. Ambiguous pricing or inconsistent inventory cause immediate disqualification.”
  • You understand that authority is a first-class product problem, not an implementation detail. Of the 72–79% of enterprises who test agentic systems, only 1 in 9 reaches production scale. The failure mode is rarely model quality — it is organizational authority. If you treat permissions, audit logging, and revocation as tail-end engineering, you will join the 8 in 9.
The market context that makes this urgent: the agentic AI market crossed 9Bin2026andisontrackfor9B in 2026 and is on track for 52.62B by 2030 (46.3% CAGR). Gartner forecasts 40% of enterprise applications will embed task-specific agents by year-end 2026. The window to establish vertical authority before consolidation is measured in quarters, not years.

Steps

Step 1 — Pick a vertical narrow enough to own

The vertical winners already named — Harvey (legal), Sierra (customer service), Hippocratic (healthcare clinical), Abridge (clinical documentation) — are no longer open. They have funding, category definition, and a head start. The open verticals are the ones that look boring from a distance. The clearest example:
“Veterinary practice management — $2.1B market. ‘Gets none of the healthcare AI investment because it sits in an unglamorous corner of the market — which is exactly why it is still open.’”
The same logic applies across several other underserved verticals:
  • Construction subcontractor compliance — $12B annual software spend, highly fragmented, compliance requirements vary by state and municipality
  • Real estate investment analysis — mid-market operators lack the data infrastructure that institutional players take for granted
  • Financial services regulatory compliance — high-stakes, high-frequency document work with significant carrier-specific variation
  • Supply chain disruption intelligence (mid-market) — enterprise players have this; the mid-market does not
  • Legal contract review for SMB — the $15B e-discovery market is dominated by enterprise tooling; SMB is structurally underserved
Prophetic is an instructive example of vertical depth: a zoning intelligence agent trained on 20,000+ US municipal codes. That is not a general-purpose AI agent with a zoning prompt. It is a product that required collecting, cleaning, and encoding a dataset that no horizontal platform will ever build.

Step 2 — Encode the workflow depth

The 3x growth and 2–4x pricing premium that vertical SaaS companies command over horizontal competitors come from one source: workflow depth. A vertical product encodes the specific compliance requirements, data schemas, terminology, and edge cases of a single industry in a way that a horizontal platform structurally cannot. Healthcare prior authorization is a useful case study because it illustrates the failure mode of generic approaches:
“Healthcare prior authorization — $35B annual administrative waste. ‘Generic AI fails on carrier-specific requirements.’ Each insurance carrier has unique workflows.”
$35B in annual waste is a large number. But the reason generic AI fails here is precisely the reason a vertical specialist wins: every insurance carrier has unique workflows. An agent that handles Blue Cross Blue Shield prior auth is not the same agent that handles United or Cigna. The product that wins is the one that encodes carrier-specific workflows at the level of detail that a trained human specialist currently carries in their head. The right entry point: one carrier, one procedure category, full automation. Expand from there.

Step 3 — Build the authority model on day one

The bottleneck in agentic AI has shifted. For most of 2023 and 2024, the primary question was whether models were capable enough. That question is largely settled.
“The bottleneck has shifted from model capability to the org’s ability to grant agents safe-but-real authority.”
72–79% of enterprises test or deploy agentic systems. Only 1 in 9 reaches production scale. The same models that power those POCs are the same models available at production scale. The failure is organizational: permissions, audit trails, exception handling, compliance review, change management. Practical authority-model requirements for vertical agents:
  1. Scope authority explicitly at the product layer. Do not make enterprise buyers figure out what your agent can and cannot do through trial and error. Ship a clear, auditable permissions model on day one. Make it easy to grant and revoke.
  2. Build the paper trail first. Every action your agent takes should be logged in human-readable terms, not just machine logs. 65% of enterprises cite scaling as their top challenge; 46% cite integration. Both reduce to: “we cannot explain to our compliance team what the agent did and why.”
  3. Design for the oversight spectrum. 34% of enterprise deployments currently use “let it rip” oversight (agents act first, humans review afterward). 78% plan to increase agent autonomy. The product that wins is not optimized for either extreme — it is the one that makes it easy to dial autonomy up or down based on action type and risk level.

Step 4 — Make your product agent-discoverable

The frame shift for builders is stark:
“The era of optimizing for eyeballs is ending. The era of optimizing for API calls is beginning.”
Human buyers search, read, and compare. Agent buyers query APIs, parse structured responses, and make decisions algorithmically. A product optimized for human discovery (SEO, UI polish, onboarding flow) and a product optimized for agent discovery (clean API surface, deterministic pricing, machine-readable inventory) are not the same product. Concrete requirements for agent-discoverability:
  • Deterministic pricing. No “contact sales for pricing” as a default. Cooper: “Ambiguous pricing or inconsistent inventory cause immediate disqualification.”
  • Machine-readable inventory and availability. Agents do not negotiate ambiguity the way human buyers do.
  • Trust signals that an algorithm can read. Third-party reviews and consistent brand data influence algorithmic recommendation weight.
  • Mandate-compatible authorization surfaces. See Step 5.

Step 5 — Align with the emerging mandate primitive

Google’s Agent Payments Protocol (AP2) is the most concrete technical response to the authority bottleneck published so far:
“AP2 introduces ‘mandates’ — digitally signed, portable, verifiable, revocable statements defining what an agent is permitted to do (create a cart, complete a purchase, manage a subscription). AP2 is an extension of A2A and the Model Context Protocol (MCP).”
A mandate is not a prompt instruction. It is a cryptographically signed, revocable authorization that travels with the agent and can be verified by any system the agent interacts with. The design solves a core problem: how does a third-party service know that the AI agent making a purchase on behalf of a user actually has permission to do so? AP2 launched with Coinbase and 60+ partner organizations. Stripe shipped Agent Toolkit. Some brands report 10% of revenue already coming from AI agents. AP2 is one of three competing protocols (alongside ACP and x402), and the standard is not yet settled — but the direction is clear: agent authority will be governed by explicit, machine-verifiable grants, not implicit prompt context. Builders who design their products around this now — exposing mandate-compatible authorization surfaces — will have a structural advantage when the standard converges.
For context on agent orchestration infrastructure and how AP2 fits into multi-agent architectures, see the adjacent entry on cloud agent orchestration without local compute limits.

Why these steps in this order

Step 1 (pick a vertical) comes first because every subsequent step depends on knowing which workflow you are encoding. You cannot build workflow depth, an authority model, or an agent-discoverable surface without a specific vertical to anchor them. Step 2 (workflow depth) comes before authority and agent-discoverability because the workflow IS the moat. A clean API with no proprietary workflow is not a defensible product. The vertical depth is what generates the 2–4x pricing premium and what keeps horizontal platforms structurally unable to compete. Step 3 (authority model) comes before Step 4 (agent-discoverability) for a load-bearing reason: an agent-discoverable product without an authority model is a liability. If agents can find you, transact with you, and operate inside your product, you must already have answers to “who authorized this action,” “can it be revoked,” and “what is the audit trail.” Reversing this order is the most common path into the 8-of-9 failure rate. Step 4 (agent-discoverability) and Step 5 (mandate compatibility) are paired — agent-discoverability without mandate-compatible authorization solves the discovery problem but not the trust problem. Cooper is explicit that trust functions as a ranking signal in agent commerce; a product that agents can find but cannot trust to honor authorization scopes will lose to one that handles both.

Common pitfalls

The 1-in-9 production scale rate is the most important benchmark in this space. The failure modes are consistent.
  • Scope too broad at launch. A prior auth agent that tries to handle all carriers at once fails because the carrier-specific edge cases multiply faster than the team can resolve them. The 1 in 9 that succeeds picked one workflow and did it completely.
  • No explicit authority model. The POC runs under an engineer’s credentials with elevated permissions. When legal and compliance review the production deployment, they have no way to scope the agent’s authority to an acceptable level. The project stalls indefinitely.
  • Integration assumptions that do not hold. 46% of enterprises cite integration as a top challenge. The vertical agent that wins has done the painful work of connecting to the legacy systems in its target vertical — the EHRs, the carrier portals, the municipal databases — rather than assuming buyers will manage the integration layer.
  • Ambiguous pricing for agent buyers. As Cooper noted, ambiguous pricing causes immediate disqualification in agent-mediated commerce. If your pricing requires human negotiation, you are building a product that agent buyers cannot purchase.
  • Building for everyone. “Building for everyone instead of someone specific is a common pitfall — generalist solutions lose to vertical specialists every time.”
The repeating pattern across vertical AI winners: pick one narrow niche, dominate it, then expand. Generalists lose to vertical specialists every time.

When this approach fails

This playbook is the wrong approach in three specific situations. You are building infrastructure, not an end product. Horizontal infrastructure plays — agent orchestration, model routing, observability — operate under different dynamics. Their moat is platform-level network effects, not vertical workflow depth. The “pick a narrow niche” advice does not apply to infrastructure positioning. Your vertical is one where horizontal already won. A small number of verticals — generic enterprise document processing, broad-purpose chat — have horizontal incumbents with enough scale and data that a vertical specialist cannot displace them on workflow depth alone. Pick a different vertical, or pick a deeper sub-niche within the same one. You cannot stomach a 12–18 month focus on a single carrier, jurisdiction, or workflow. The vertical AI position rewards depth and punishes premature expansion. If your strategy or runway requires showing breadth before depth, this is not the right framing — and a horizontal play with explicit carrier add-ons later may be a better fit despite the structural disadvantages.

What we learned

The three-force convergence is not a prediction. Each force has named companies, published financials, and executive-level public statements behind it. The builders who will capture disproportionate value in this cycle are the ones who pick a vertical narrow enough to own, build an authority model explicit enough to survive enterprise compliance review, and design an API surface clean enough for agent buyers to transact programmatically. The unglamorous verticals — veterinary practice management, construction subcontractor compliance, prior authorization for a single carrier — are still open. They are open because most founders are looking for the glamorous version of this opportunity. That is the opportunity. The 300-unicorn prediction from TheFounderVC’s Mia Lewin will look prescient or conservative depending on how quickly the authority bottleneck gets solved at the infrastructure layer. Either way, the vertical AI IPOs she expects within three years will be companies that started building their domain-specific moats in 2025 and 2026 — and that treated agents as customers from day one.

FAQ

Q: What if my vertical isn’t on the underserved list? The list (veterinary, construction subcontractor compliance, real estate investment analysis, financial services compliance, supply chain mid-market, SMB legal contract review) is illustrative, not exhaustive. The pattern is “unglamorous, fragmented, with carrier-specific or jurisdiction-specific edge cases that generic AI cannot encode.” If your vertical fits that pattern and lacks well-funded incumbents, it qualifies. If your vertical has a Harvey-scale incumbent, the position is to find a sub-vertical they ignore (e.g. SMB law firms below the scale Harvey targets). Q: Do I need to wait for AP2 to standardize before building? No. AP2 is one of three competing protocols (alongside ACP and x402). Designing your authorization surface around AP2 today gives you a structural advantage if it wins; if a different standard wins, the abstraction work translates because all three protocols solve the same problem (signed, revocable, machine-verifiable agent authority). The risk of designing for none of them is much higher than the risk of picking the wrong one early. Q: Should I optimize for human buyers first and add agent-buyer support later? This is the most common reasoning trap. Some brands already report 10% of revenue from agent buyers, and infrastructure (Stripe Agent Toolkit, AP2, agentic checkout pilots at Google, Amazon, Meta) is shipping in 2026. The cost of designing for agent discoverability from day one is small. The cost of retrofitting an agent-buyer-friendly surface onto a product designed exclusively for human buyers is structural — it requires changing pricing models, API contracts, and trust-signal architecture simultaneously. Q: How narrow is “narrow enough” for the initial vertical? Narrow enough that a single domain expert can validate every edge case in your training data and product workflow. For prior auth, this means one carrier (not “healthcare”). For construction compliance, this means one US state’s subcontractor regulations (not “construction”). For legal contract review, this means commercial leases for SMB firms (not “contracts”). Expansion happens after dominance, not in parallel with it. Q: How do I sell to enterprise buyers when authority is the gating concern? Lead with the audit trail, not the capability demo. Enterprise compliance teams are not impressed by what your agent can do — they are blocked by what they cannot explain to their auditors. Show the human-readable action log, the explicit permissions scope, and the revocation flow before you show the workflow automation. The 1 in 9 that reach production scale do this; the 8 in 9 that stall do not. Q: What does “B2A-native” actually mean in product terms? A B2A-native vertical agent is not an existing SaaS product with an AI layer added. It is a product designed from the ground up around three assumptions: (1) agent buyers are a primary customer segment, not an afterthought; (2) vertical depth is the moat, not model quality; (3) authority is a first-class product problem, not an implementation detail. Each of these manifests as concrete product decisions — pricing, API surface, audit logging, permissions model — that are difficult to retrofit later.