The a16z Show

Seema Amble & Steven Sinofsky on why software isn't losing its head

Seema Amble & Steven Sinofsky· a16z Enterprise Partner & Board Partner at Andreessen Horowitz (a16z)
·~61 min·English·a16z
AgentsBusiness StrategyAI Company
TL;DR

a16z's Seema Amble and Steven Sinofsky argue that 'headless' software doesn't kill the incumbents — because the value was never the UI but the codified business logic, exceptions, and institutional memory underneath, and agents change how you reach that value, not who has to own it.

01Core Mental Model

The Head Comes Off, the Body Stays

'Headless' doesn't mean dead — it means the UI became optional while the data and logic underneath stayed exactly where the value always was.

In an agentic world, do you actually need that? The data, the logic, everything stored below it is really where the value is.

— Seema Amble, The a16z Show
Key Insight
Salesforce's 'Headless 360' was largely a rebrand of APIs that had long existed — the real news isn't a new product but an admission that the interface layer, long the moat, is now the disposable part.
02What Agents Actually Do

Look Up, Do, Analyze

Before asking whether an agent can replace software, ask which of three jobs it's doing — looking up, doing, or analyzing — because the difficulty and the danger differ wildly.

it's also where hallucination really is a huge issue because if you're going to go and analyze something, you actually need a way to verify that everything every step of that analysis was correct.

— Steven Sinofsky, The a16z Show
Key Insight
The industry conflates 'headless' and 'agentic,' but they sit on different rungs: most shipped 'headless agent APIs' are just lookup with a nicer interface, while the genuinely transformative mode — analyze — is also the one whose output you can't trust without verifying every step.
03Why Software Doesn't Die

Stickiness Is Discovered, Not Designed

No product manager ever engineered stickiness — it emerges from usage, money, and muscle memory, and you only find out what it is when a customer threatens to leave.

the most sticky thing you could do is actually collect money from a customer. And if you're collecting money, it turns out it's really really hard for them to stop sending you money.

— Steven Sinofsky, The a16z Show
Key Insight
Sinofsky's tell: the way you find what's sticky is to have a rep listen when a customer threatens to rip you out — do that across a few accounts and the real reason surfaces, and it's usually something arcane (Outlook's calendar delegate access, not its email) that no roadmap ever prioritized.
04The Deepest Moat

You Can't Vibe-Code SAP

Enterprise software isn't a database with a UI bolted on — it's the company's business rules codified, which is why ripping out SAP would dissolve the company that runs on it.

Misconception right now is that you can just have you know Postgress database and APIs and then bam like you can replace SAP. That's like absolutely not true.

— Seema Amble, The a16z Show
Key Insight
Sinofsky's sharper version: Ford, Toyota, and GM all 'just make cars' — what differentiates them is which customizations they chose inside SAP. When a Goldman banker told a young Sinofsky they made more money from Excel than Microsoft did, the point was identical: the edge lives in the add-ins and models poured into the tool, not the tool.
05The Long Tail

Everything Interesting Is an Exception

The data in the fields is the easy part; the value and the difficulty both live in the exceptions — the edge cases, policies, and context that were only ever in someone's head.

almost everything interesting in an enterprise is an exception.

— Steven Sinofsky, The a16z Show
Key Insight
This is why 'context graphs' matter: the context IS the exceptions, and they accrue at the pace of a sales cycle — one at a time, slowly — so capturing them is a long observation problem, not a one-shot data export.
06The Anti-Doomer Argument

The Long Tail Got Longer

Automating the mundane doesn't shrink the work — it frees capacity that immediately gets spent inventing new, higher-order work nobody did before.

the long tail got no shorter. It just got longer in a different way.

— Steven Sinofsky, The a16z Show
Key Insight
The doomer error treats work as a fixed pie of N people times M software. The counter: productivity invents scenarios. Amazon killed the miserable phone-return experience, then had to build an entire backend to figure out why wrong items ship at all — the mundane task vanished and a bigger analysis job appeared in its place.
07The Middleware Trap

Nobody Wants to Be Middleware

Every vendor resists being reduced to a dumb database beneath a 'benign' orchestration layer — so that middle layer is always the least stable place to build.

no software wants to be disintermediated by some other layer above it.

— Steven Sinofsky, The a16z Show
Key Insight
Two forces squeeze the middle: incumbents below refuse to be commoditized (Workday deliberately makes clean extraction hard; SAP just ships the ecosystem's best features itself), and your own stack inherits the fragility of its weakest dependency. The clean-API dream is an engineer's aesthetic the market keeps rejecting.
08The Startup Playbook

Aim for the Middle

The dumbest move in a platform shift is attacking a category head-on; the winning move is to slip between two incumbents who will only ever bolt AI onto the side.

your opportunity in a startup is to just look at two big players who are bolting AI onto the side and exposing some existing API as an agent or whatever and just aim for the middle and do things in in the new way.

— Steven Sinofsky, The a16z Show
Key Insight
The historical rhyme: the web beat client-server despite a trillion dollars invested in the old model — not by doing what client-server did (it did none of it) but by implementing the concept anew. Amble's corollary: the richest gaps now sit between functions inside a company — the handoffs — that AI can finally bridge.