Distribution Beats Capability: Why Incumbents Hold the Real Advantage in the Agent Era

Table of Contents

• The Bet Everyone Gets Wrong
• What Distribution Actually Means Here
• Why Capability Commoditizes Faster Than Distribution
• The Four Layers of an Incumbent's Distribution Moat
• Where the Distribution Advantage Actually Leaks
• How Agent-Native Startups Route Around the Wall
• What This Means for Buyers and Builders
• Insights Most People Overlook
• References

The Bet Everyone Gets Wrong

Spend a week in agent demos and you start to notice the pitch is always the same shape: ours reasons better, ours hallucinates less, ours handles the edge case yours fumbles. Capability theater. It's a reasonable thing to compete on if you believe enterprise software gets bought the way a benchmark gets scored, best number wins.

It doesn't. Enterprise software gets bought the way a renewal gets signed: by the person who already has the budget line, talking to the vendor who already has the master service agreement, about a product that already touches the data. The agent that wins the deal is frequently not the most capable one. It's the one that shows up inside a tool the buyer already pays for, requires no new security review, and gets switched on with a checkbox.

That's the uncomfortable thesis of this corner of the GaaS cluster: distribution beats capability. A mediocre agent with default distribution will, more often than not, beat a brilliant agent that has to be discovered, evaluated, procured, integrated, and trusted from a cold start. Marc Andreessen's old line that the number one company-killer is lack of market, not lack of product, written years before agents existed, turns out to describe the agent wars almost perfectly.

If you've read the SaaS is dead steelman and rebuttal elsewhere in this beat, treat this article as the counterweight. The incumbents are not as doomed as the disruption narrative insists, and the reason is unglamorous.

What Distribution Actually Means Here

Distribution is a slippery word, so let's be precise. In the agent context it isn't "we have a sales team" or "we run ads." It's the sum of everything that makes it easy for an agent to reach a user and start doing work, without the user having to go find it.

Concretely, an incumbent's distribution is:

• The installed base. Salesforce already has the CRM seats. ServiceNow already runs the IT workflows. Microsoft already owns the inbox and the document. The agent doesn't need to acquire those users; it inherits them.
• The system of record. The agent that lives where the data already is doesn't have to beg for an integration. This is the spine of the system-of-record vs. system-of-action battle, whoever holds the record holds the easiest path to action.
• The procurement relationship. A signed contract, a vetted vendor, an existing security posture, a renewal cadence. Adding an agent to an existing vendor is a line-item change. Adding a new vendor is a project.
• The default surface. The button in the UI the user already stares at all day. Default placement is worth more than feature parity, because most users never leave the defaults.

Notice none of this is about how smart the agent is. It's about proximity, to the user, to the data, to the budget. Capability is what the agent does once it arrives. Distribution is whether it arrives at all.

Why Capability Commoditizes Faster Than Distribution

Here's the structural asymmetry that makes the thesis hold. Capability is converging; distribution is entrenching.

Frontier model quality is improving fast and diffusing faster. The gap between the best model and the third-best model, measured on the tasks that actually matter to enterprise workflows, narrows every quarter, and open-weight models trail the frontier by months, not years. McKinsey's research on the economic potential of generative AI makes the point indirectly: the value isn't in the model, it's in wiring the model into real work. The model is the commodity input. The wiring is the moat.

So if you're an agent startup betting on a capability lead, you're betting on a depreciating asset. Whatever clever scaffolding, fine-tune, or eval harness gives you an edge this quarter is reproducible next quarter, and the quarter after that the base model just does it natively and your edge evaporates. This is the thin wrapper panic in a sentence: capability moats erode from underneath because the foundation model keeps rising.

Distribution does the opposite. Every quarter an incumbent's agent is the default, switching costs compound. The workflows get tuned to it. The data accumulates inside it. The org chart reorganizes around it. A capability lead is a sandcastle at the tide line. A distribution lead is a seawall someone poured concrete into for ten years.

This is exactly why the smart incumbents are racing to embed agents now, even when their agents are visibly worse than the startups'. They understand the clock. Get to "good enough and already here" before the startup gets to "excellent but you've never heard of it," and the game is largely over. That's the logic behind every legacy SaaS adding agents move you're watching unfold, even when it does look like lipstick.

The Four Layers of an Incumbent's Distribution Moat

It helps to disaggregate the moat, because each layer leaks differently and a startup needs to know which one it's actually fighting.

Layer 1: The Identity and Access Layer

The agent that runs as you, inside your authenticated session, with your permissions already provisioned, skips the single hardest step in enterprise software: getting access. An incumbent agent inherits SSO, role-based permissions, audit logging, and the org's entire identity graph for free. A new vendor's agent has to negotiate every one of those, and in a regulated enterprise that negotiation can take quarters. Identity is the quietest and stickiest layer of distribution.

Layer 2: The Data-Gravity Layer

Data has gravity; agents fall toward it. The agent co-located with the system of record reads and writes without a brittle integration in between. This is the heart of the data moat agents can't easily cross, and increasingly the site of open conflict, incumbents are starting to gatekeep the API access that outside agents need, which the data-access wars cover directly. When the incumbent can throttle your integration, your superior capability runs on stale or partial data, and capability without data is a parlor trick.

Layer 3: The Procurement and Trust Layer

Enterprises don't buy agents; they buy vendors they've already cleared. An existing vendor adding an agent triggers a change order. A new vendor triggers security review, legal review, data-processing agreements, and a budget fight. The friction differential is enormous and almost entirely invisible in a product demo. It's also why procurement is changing, but slowly, and the incumbent benefits from every day the old motion persists.

Layer 4: The Default-Surface Layer

The last layer is the cheapest to describe and the most underestimated. Users live in defaults. The agent that appears as a button in the tool already open on the screen gets used; the equally-good agent in a separate tab gets forgotten. This is the agents as the new UI layer dynamic from the incumbent's side: own the surface, own the usage, regardless of what's underneath.

Where the Distribution Advantage Actually Leaks

If incumbents always won, this beat would have one article. They don't, and the leaks are specific. Honesty about them is what separates analysis from cheerleading.

Distribution is workflow-specific, not universal. An incumbent's moat is deep inside its core workflow and shallow everywhere else. Salesforce's distribution into sales agents is formidable; its distribution into, say, a security-operations agent is roughly zero. The great unbundling happens precisely at the seams between an incumbent's strong and weak surfaces, agents pick off the workflows the suite hosted but never owned.

The buyer is changing, and new buyers have no loyalty. When the budget shifts from a software line to a labor line, the shift from software budgets to labor budgets, the person approving the purchase changes too. A COO buying outcomes has no relationship with your CRM vendor's account rep. Distribution built on the old buyer doesn't transfer to the new one. This is the single biggest crack in the wall.

Incumbent distribution can be hostage to incumbent incentives. A vendor charging per seat cannot enthusiastically ship an agent that eliminates seats. The incumbent's dilemma, cannibalize seats or lose to startups, means the incumbent often throttles its own agent's ambition to protect the existing model. That self-imposed handicap is the oxygen a startup breathes. The startup ships the agent that actually does the whole job; the incumbent ships the one that does just enough to defend the renewal.

Default placement only matters if the user opens the tool. When an agent replaces the reason to open the tool, when the work moves to chat, to email, to a browser agent, the default surface stops being a surface anyone visits. The browser-agent threat to web SaaS is exactly this: distribution that depended on a destination collapses when the destination becomes optional.

How Agent-Native Startups Route Around the Wall

The mistake startups make is fighting distribution with capability, charging the seawall with a better catapult. The ones that win route around it. A few patterns are emerging.

Pick a workflow the incumbent can't or won't serve. Go where the incumbent's distribution is shallow and its incentive is conflicted. Vertical, regulated, cross-system workflows that no single suite owns are the soft targets, the terrain of vertical SaaS vs. vertical agents. If the incumbent has to cannibalize a seat business to compete, you have time the moat can't buy back.

Sell the outcome to the new buyer. Don't sell software to the software buyer; sell a completed job to the labor budget. When you're priced as opex labor rather than a tool, the CFO reframing of agents as opex labor, you bypass the procurement relationship the incumbent owns entirely, because you're not in that budget at all.

Become the distribution. The most durable counter is to make yourself the surface, the agent layer that sits above multiple systems and owns the customer relationship, per the agent layer's claim on the customer relationship. If you become the thing the user opens, the incumbent's default-surface advantage inverts: now they are the commodity backend and you hold distribution.

Ride someone else's distribution. Agent marketplaces and app stores are a borrowed-distribution play, see app stores for agents. The catch, covered in platform risk, is that borrowed distribution can be revoked, repriced, or competed with by the platform itself. It's a real channel, but you're a tenant, not an owner.

What This Means for Buyers and Builders

If you're buying agents, stop scoring demos like benchmarks. The agent that's marginally better in a bake-off but requires a new vendor relationship, a fresh security review, and a separate data integration may cost you more in friction than it ever returns in capability. Weight time-to-value and integration depth as heavily as raw quality, and be honest that the incumbent's "good enough and already here" is often the rational buy, even when it stings.

If you're building, internalize that your capability lead is a clock running down, and spend it buying distribution before it expires. Use the window where you're genuinely better to embed so deeply, into a workflow, a data position, a customer relationship, that by the time the incumbent's agent catches up on capability, you've become the default they have to dislodge. Capability gets you in the door. Distribution is what keeps you in the building. The incumbents already learned this lesson, which is precisely why they're moving so fast, and so cynically, to switch their agents on.

Insights Most People Overlook

• The incumbent's worst enemy is its own pricing model, not the startup's capability. A vendor whose revenue is seat-count cannot ship the agent that does the ten-person job, because it's volunteering to shrink itself. Startups don't beat incumbents on intelligence; they beat them on the incumbent's unwillingness to be intelligent enough to cannibalize its own seats. The moat has a self-inflicted hole shaped exactly like the business model.

• Distribution and capability trade places when the agent becomes the destination. Default-surface advantage is conditional on the user opening the tool. The moment work migrates to an always-open agent, chat, email, browser, the incumbent's "default button" sits on a page no one loads. Distribution built on a destination is contingent distribution, and most incumbents are underwriting a moat that assumes their tool stays the place work happens.

• Data gatekeeping is a tell, not a flex. When an incumbent starts restricting API access to outside agents, the instinct is to read it as strength. Read it the other way: a vendor confident in its agent's capability lets rivals connect and still wins on quality. Gatekeeping is what you do when you suspect the rival agent is better and your only durable advantage is the data pipe. The restriction is a confession.

• "AI-native" is a distribution claim disguised as a capability claim. When everyone says they're AI-native, the phrase stops describing the model and starts describing trust positioning, a bid to be the default agent vendor, not the most capable one. It's marketing aimed at the procurement layer, not the engineering one. Decode it as a distribution play and it makes far more sense.

• The riskiest distribution is the kind you rent. Marketplaces and platform channels feel like distribution, but a landlord who can raise the rent, evict you, or open a competing shop next door isn't giving you a moat, they're giving you a lease with a termination clause. The startups that mistake borrowed reach for owned distribution are building castles on someone else's land, and the platform's roadmap is the eviction notice they haven't read yet.

References

• The Only Thing That Matters, Marc Andreessen, Andreessen Horowitz (a16z)
• The Economic Potential of Generative AI: The Next Productivity Frontier, McKinsey & Company