a16z Podcast

David Pawlan on why the invisible agent wins

David Pawlan· Creator of Assistant Bench at Assistant Bench
·~50 min·English·a16z
AgentsAI CompanyBusiness Strategy
TL;DR

<strong>The winning assistant is invisible and proactive</strong> — it saves you money in the background, and asks permission only before consequential changes.

01The Landscape

A Scoreboard for a Space With No Map

<strong>Pawlan's Assistant Bench tracks 122 agents by use case</strong>, across consumer and B2B, because a category moving this fast has no map yet.

So far to my knowledge, there are 122 different ones across all the categories.

— David Pawlan, a16z Podcast
Key Insight
Over 100,000 site visitors, plus inbound from “basically every single founder,” is the tell: the market itself cannot yet say who is ahead, so a neutral scoreboard became the first product the category needed.
02Core Thesis

The Winning Assistant Is Invisible

<strong>The winning assistant is invisible</strong> — consumers do not want to be 10% more efficient, they want the work done and reported back like a great employee.

The general population does not care about being 10% more efficient.

— David Pawlan, a16z Podcast
Key Insight
His bar is not capability, it is absence: an agent that demands attention has already failed — which inverts how most AI products are sold, with feature lists and dashboards. The surface he describes has almost none.
03Winning Use Cases

Cost-Savers, Not Cost-Makers

<strong>The breakout use cases are cost-savers, not cost-makers</strong> — HSA refunds, flight-price credits, and a Grok-bot sprinkler wired to the weather that halved a water bill.

He hooked up his Grok bot to his sprinkler system at his home and connected it to the weather system.

— David Pawlan, a16z Podcast
Key Insight
“Free money” reframes savings as gains to beat consumer apathy. The real unlock is not the agent’s intelligence but its willingness to grind through bureaucracy — receipts, refund forms, credits — that no person will do for themselves.
04Defensibility

Proactivity Is the Moat, Not Personality

<strong>Proactivity is the moat, not personality</strong> — a personality is a configurable soul file anyone can copy, while useful unprompted action is what can set the winner apart.

I think there is massive defensibility around proactivity.

— David Pawlan, a16z Podcast
Key Insight
If personality is a prompt and proactivity is a behavior, the moat moves from how the agent talks to what it dares to do on its own — a capability gap, not a branding one, and far harder for a fast-follower to close.
05Trust

The Fine Line You Only Cross Once

<strong>Cross the trust line once and you lose the user</strong> — the agent should act on its own when nothing changes on your side, but ask first before anything consequential.

if you cross that line once, you lose all the trust with your user.

— David Pawlan, a16z Podcast
Key Insight
The trust asymmetry is sharp: one wrong autonomous action can undo a lot of goodwill at once. So the winning agent is less the most capable than the best at telling apart what needs no action on your side from what changes an action on your side — and asking only for the second.
06Multiplayer

The Best Group Agent Says Nothing

<strong>The best group-chat agent stays silent</strong> — it listens, takes notes, and only ever pings one person privately with an action item.

that agent is silent, and it's purely a listener.

— David Pawlan, a16z Podcast
Key Insight
The intrusive feeling of an agent in a group chat is a social problem, not a technical one. Pawlan’s favorite design removes the agent from the conversation entirely: it earns its place by being useful to one person at a time, never by taking the floor.
07New Infrastructure

The Agent-to-Agent Economy

<strong>Pawlan expects a whole economy for agent-to-agent interactions</strong> that do not exist today — agent emails, agent phone numbers, and a new security layer.

agent booking from agent and how does that redefine the entire service industry as a whole?

— David Pawlan, a16z Podcast
Key Insight
Pawlan’s claim is structural: once both ends of a booking are agents, the service industry built on humans transacting with humans has no obvious shape left. That vacuum is why he expects a Cambrian wave of startups — and a new security layer — to define the agent-to-agent plumbing.
08Economics

Why Charge When the Giants Are Free?

<strong>Most agents charge, yet the two that dominate are free</strong> — leaving Pawlan asking how a paid long-tail startup can possibly win.

why are they charging money when the big dogs are giving it to you for free?

— David Pawlan, a16z Podcast
Key Insight
Pawlan’s own numbers frame the puzzle he poses: 65 of 122 agents charge, yet the two dominant ones are free. A paid startup then carries the burden of proof — it needs a clear reason to pick it over a free default before pricing ever helps.