The a16z Show - Is Software Losing Its Head?

Episode Date: July 7, 2026

Seema Amble, Steven Sinofsky, and Elena Burger unpack one of the biggest questions facing enterprise software: what happens when AI agents become the primary users of software instead of humans? The conversation explores the rise of "headless" software, why APIs and agentic workflows are reshaping enterprise applications, and whether traditional SaaS products are becoming systems of record rather than systems of engagement. They discuss Salesforce's Headless 360 announcement, MCP, enterprise software architecture, and why AI may fundamentally change how businesses interact with their data. Along the way, they examine what actually makes enterprise software sticky, why replacing systems like SAP and Salesforce is harder than it appears, and where startups have the greatest opportunity as AI reshapes the software stack.   Resources: Follow Seema Amble on X: https://x.com/seema_amble Follow Steven Sinofsky on X: https://x.com/stevesi Follow Elena Burger on X: https://x.com/VirtualElena Related Reading Is Software Losing Its Head?https://a16z.com/is-software-losing-its-head/ The Death of Software? Nah.https://a16z.com/death-of-software-nah/ Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

There are many things that made software sticky, but a lot of it had to do with the fact that it was built around, like, the way a human interacts. In an agendic world, do you actually need that? The data, the logic, everything stored below it is really where the value is. There's this wild underestimation about, like, you could vibe code your way into enterprise software. Larry Ellison at Oracle, he went on a rant about how enterprise software was so stupid because everybody customized it.

The minute you automate the most mundane thing and think you have it all squared away, whole new things appear. Misconception right now is that you can just have, you know, postcast database and APIs and then bam, like, you can replace SAP. And that's like absolutely not true. That piece around the logic and everything else that is in captured in SAP is way, way more important than the fact that like, oh, this data just happens to be in this database. One of the things that happens in technology shifts is nobody understands exponential when it's, happening. The biggest opportunity right now is... For decades, enterprise software has been built around one assumption. Humans are the primary users. But what happens when AI agents become the ones reading data, updating records, and

completing workflows? That shift raises a much bigger question than whether software gets a new interface. It challenges how enterprise software is built, where value lives, and what makes platforms like Salesforce and SAP so difficult to replace. In this, episode, Sema Amble, Steven Sinovsky, and Elena Berger discuss headless software, AI agents, enterprise architecture, and why the next generation of software may look very different from the SaaS products we've used for the last 20 years. Welcome to the A16Z podcast. I'm here with Seema Amble, a partner here on the enterprise team, and Steven Sinovsky, who is a board partner at A16Z, as well as a former member of Microsoft, a friend of the firm.

And here we are today to talk about a piece that Seema wrote about a month ago called Is Software Losing Its Head? And I'll let Seema talk about it in her own words. But this piece was written a couple months ago. Salesforce announced that they would be going headless. And today we're here to kind of discuss what does that mean? What does that mean for the future of SaaS products, the future of kind of software more generally? So, Seema, can you just walk me through first what Headless Software means and explain kind of what changes it introduces?

Yeah. So Headless Software is not a new term, but I think has really risen in the public domain of interest in a topic that people are talking about. One of the interesting news points has been Salesforce making this announcement. They were launching Headless 360, which was really in classic Salesforce motion history. a marketing announcement more than anything else, but it does capture, it's an acknowledgement of what's happening, which is traditional software had been built around humans accessing it, and it was workflow to capture data,

and we could talk more about what that meant. In an agendic world, do you actually need that? The UI doesn't matter because the agent isn't accessing the software via the UI. We could unpack whether the UI matters or not, but in the idea of it being headless is the data, the logic, I mean, stored below it is really where the value is, not just the workflow software that's being tracked at the top.

And this was announced a couple months ago. I'm curious in the past couple months, what have we seen? I mean, you wrote, even in the piece, kind of at the beginning, you were a little bit funny, maybe, and you said, is this really even that big of an announcement? Is this sort of a rebrand of APIs that they had kind of already made available? So does this feel like a significant change? this may be more of a branding exercise, and kind of like what have we even seen in the past couple months since we've been able to kind of observe what changes have really happened.

So I'll separate it out into the Salesforce context and then the broader context, which I think the broader context is more interesting. In the Salesforce context, probably not that interesting. I think Salesforce, I think rightfully, again, it was acknowledging a shift that's happening the market. From what I could tell, nothing actually changed. Their 360 product was the same APIs that had always been exposed now rebranded as their 360 product. And APIs have always I should say always, but for a long time it existed. But I think the broader trend here is, you know, Salesforce among others are thinking about how they build themselves for the agentic world. And so if an agent needs to access the data in a CRM like Salesforce, what are they doing it? Are

they doing it via the UI or are the API? And that's the Salesforce is saying, okay, hey, we know what is changing, that there are agents who are needing to access the data. Let's offer a headless version for them to interact with the data versus going via the UI. That said, again, I don't think it actually changed in the Salesforce context, but Salesforce isn't the old one. Another example is Notion has a headless product. And actually, I think that makes even more sense because it's much hard. Like, I think many users of Salesforce are probably less technically adapt, less likely to be building their own agents, although there are many, many more people who are doing that at Salesforce. Notion users, all things,

being equal are more, I'd say, tech savvy and more agentic as builders. And I think Notion is one of many other companies that is also trying to figure out, okay, what is it that I offer and what APIs do I expose? I think Stephen will talk more about MCP. And again, I think a lot of this is also getting caught up a little bit in nomenclature and like, okay, what are we calling things? And I think that's one thing, but I think the broader trend around how agents access systems of record, I think, is the bigger point. Yeah, yeah. And from my understanding, it's also, I mean, it could also apply to something as simple as a chatbot. You know, it's not necessarily just an API or an MCP server. You could be, you know, Salesforce acquired Slack a couple of years ago. And it could be something as simple as you sort of interacting with a CRM via chatbot. Totally. And I think I read somewhere that there has been like a 300% increase in Slack buy, Slack agent usage. Yeah. Which is essentially saying that you don't need to log into the Salesforce interface to get. the data or whatever data it is. So, yeah, it's essentially, again, all these are agentic ways of accessing it versus the human

needing to go in, log the data, or from a read perspective, go back and see, okay, here's this opportunity, here's what happened, and look at themselves. And so that interface is less relevant. And Stephen, do you have anything to add here just on the kind of like definitional territory that we're covering right now or kind of this discussion? Well, sir, I mean, like, we're in definitional hell right now. part of a new wave of technology is you make up a lot of new words for things that you kind of did before.

And that's just a natural part of technology evolution. But I actually think it's super important. Like, first you have this agent, which as far as I can tell right now, is also a new word for program that takes a very long time to run and might not finish. And that's the best branding ever is to call a program that takes a really long time,

which we used to call a bug, is now like the coolest new feature ever, and it's just now an agent. But in seriousness, the most interesting way I think to think up what you're really talking about doing differently between an agent and an API is really what are you actually doing?

What is the agent itself doing? Is it looking something up? Because that's actually a pretty lightweight thing that all systems are pretty good at, And in fact, many, many of the newly announced headless agent APIs are just lookup. You basically have a new interface to the old way to look something up, which is a lot more forgiving, a lot less UI goo and stuff like that. Then there's, I want to do something. And that's where you get into very interesting issues over, well, if you do something, you have to be impersonating a specific person, you have to have their credentials.

Is it another paid seat? Is it the same paid seat? You have all these interesting enterprise software issues that come up if you actually want to cause a change to a system of record. And then there's the third thing, which is Analyze. And so Analyze is more than look something up. It's actually look up a bunch of stuff.

It often involves multiple systems. And that seems very, very tuned to an agent because you're not time-bounded. You can spend energy iterate, you can route it to different models and get different answers back and compare them. But 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. And so I think it's super interesting and important when you look at headlists and agent, which are conflated, you sort of have to figure out what you're talking about Because we're on different places in the evolution, the learning curve, and the deployment of agents relative to sort of that three-way matrix. Yeah, I think this is actually a good lead-up

into a follow-up question, which is historically what has made software sticky and how are agents starting to disrupt that? I leave that to either of you to answer. Maybe you guys can both kind of debate about that. Yeah, I'd say there are many things that made software sticky, but a lot of it had to do with the fact that it was built around like the way a human interacts, right? So the UI was sticky because, you know, a number of times you had to read and write, the frequency of access, the downstream workflows, all of the like undocumented what we call

SOPs or center operating procedures, all the stuff that happened around the software that got ingrained in muscle memory and process and then external parties, et cetera. So a CRM may be sticky because a sales rep needs to go in and out of it all the time. They're used to interacting with Salesforce. A lot of times when like new VPs of sales come into our companies, they're like, they mandate that Salesforce is there because they're used to using it. Their teams are used to using it. And then there's, you know, finance may rely on the Salesforce output for billing and

upstream marketing is going to rely on it as well. And so there's these dependencies. And these all are driving stickiness. But I think the other piece, too, is there you need one single set of truth, right? you need to know, like, an account is closed and who is working on it, and all of that needs to be logged in one place. And I think if you go from CRM to, say, an ERP or payroll, like, that absolutely has even, like, legal reasons and compliance reasons why you can't have numbers that are not being, you know, tracked as cleanly and correctly as, you know, an auditor might like, for example. So anyways, these all drove stickiness and durability because, you know, you were used to using Salesforce. The whole ecosystem was using Salesforce, and it was the default option.

Maybe, you know, there's one or two others in the market. And so I think historically those are some of the things that were driving stickiness. Yeah, I mean, those are all exactly right. I think it's important to also consider that the most sticky thing you could do is actually collect money from customer. And if you're collecting money, it turns out it's really, really hard for them to stop sending you money, and it's really hard for them to figure out what to do if they stop sending you money. And it sounds really trite, but the stickiest software is software that's getting used somewhere. Then when you dig in and try to come up with reasons, well, it just depends

on who you talk to in a company. You know, you talk to the HIPAA compliance people in some company, and they're going to tell you, this is the software you have to use because it's like the most bestest HIPAA compliant software. If you talk to the administrators, you're going to hear about onboarding new users. If you talk to the users, you're going to hear about muscle memory

and keystrokes or labor unions or whatever. And so you really want to get the software sold, and that's your fastest path to sticky. And after that, it's sort of, you know, a winner's tale over what caused it to be sticky. And in fact, the best thing about sticky is, if you're the rep for a company that you've sold something to,

and the company is threatening you like, hey, we're going to replace you, we're going to, you're just going to listen to them, and you're going to find what's sticky. And if that works like three or four times across different accounts, then you've just told the tale as to what made the product sticky. It doesn't matter what the PMs or what anybody else thought It could be some crazy arcane thing.

You know, I have stories of lots of sticky software and lots of arcane things, but anyone who's ever tried to displace Microsoft Outlook as email very quickly learned about delegate access and having calendars owned by multiple people and all of this crazy stuff. And like, I can tell you there was no meeting where we said, okay, let's figure out how to make the calendar

the sticky part of Outlook and make sure we handle recurring meeting, exception handling well. And then you go and you find out, you know, General Motors isn't going to displace like 600,000 seats because of the calendar or some crazy thing like that. And so you can,

it's really amazing in enterprise software what causes sticky and how you can actually capitalize it on it when somebody threatens to take you out of the enterprise. I think there's a really good point there. And two things. inertia is a really powerful force. And then I think the other thing is, yeah, nobody, when they're building the software

and they're at, like, thinking about, like, this rubric we put together and, like, tick, tick, tick, we got all these features that are going to do all these things. But I think the practical reality is also, as software extends its tentacles across an organization and it gets ingrained in people using it and they've been paying for it for a long time, it just seeps into how people are doing things. And that's, like, hard to rip out. Yeah. Yeah, you even talk about this in your piece, Mo, where they're sort of all.

all of these like invisible, tacit sort of understandings about how to use different products or things that are embedded both within the software but also within the people using them where it's just like it does become hard after a while to extricate yourself from whatever ecosystem you happen to be. And in fact, Stephen, I think you've even, you know, said the Saspocalypse is overblown. You've run an essay called The Death of Software, nah.

where you, you know, emphatically sort of rejected this idea. So maybe if you want to maybe just recount that piece a little bit for us. Well, I mean, Sima co-wrote a post on SAP, which is sort of the ultimate, ultimate example of sticky software. I mean, there is nothing. The only software that's stickier than SAP is behind the scenes, and it's the software that insurance companies wrote. And they wrote all this software, like, 50 years ago.

or 75 years ago. And if you ever, there's no replacing it. Like, it just, it, in fact, whenever jokes come up about, like, businesses that are looking for cobal programmers, it's to go and work on the insurance software that exists in every state of the union. And in many ways, what you're seeing,

like with one of the biggest successes to date in Strite has been somebody actually went in, and for the first time in two-generations, generations coded up the software to collect money from people, which itself had previously been an unsolved problem on the scale of insurance. Because nobody put together the tax laws for every country, every jurisdiction, every locality, every border crossing, every currency exchange. Like, it's mind-blowing that. And so now that is the most sticky, like, that is not going anywhere ever.

Like it'll be, like we'll be doing this podcast with like great grandchildren a hundred years from now talking about how sticky that experience was just like I told like, oh, you didn't know this,

but the software that runs Allstate is older than me. And it's not going anywhere. And that's because these examples are ones that codified an external force. And that external force

was the regulatory body that they embraced. The Seamus examples of SAP, like, it just codified a company. And so, like, if you take SAP out of a large automobile manufacturer, there's no automobile manufacturer left or Lamart. Like, the company just evaporates because the company is defined, not just by purchasing the software, not even by just using it,

but by how they codified the business rules into that product. I think it's a good point to double-click on because I think a misconception right now is that SAP, okay, well, you can just have, you know, Postgres database and APIs and then bam, like you can replace API, replace SAP. And that's like absolutely not true. I think partly, I mean, Stephen, I don't know if you want to elaborate and I'm happy to, but I think it's that piece around the logic and everything else that is, like, that is in captured in SAP is way, way more important than the fact.

back to that like, oh, this data just happens to be in this database. There's a reason why also SAP, like, takes, you know, multiple years to implement and it's not because, like, oh, you know, yes, the system integrators are slow and part of it, but it's, it's customized to the way that business actually operates. And I think that's, like, an important part about why you can't just obscure away the software completely from and turn it into a database plus APIs. It's just so important because this is one of the things. the things where like startups look at enterprise software and they think about it in terms of

startup scale. And so they take something like mundane like expense reporting and like, okay, well, we have 40 people and like, you know, one person can figure out expense reports for 40 people. Like you could hire a human and that and be done with it. Like literally you come back from a trip, you dump the receipts in a bucket and one human rifles through them and the expense reporting problem goes away. Or you say, oh, forget the human. We'll just all take pictures of our receipts and OCR it and categorize it and the whole thing will go away. And that's fine until you have 100,000 people in 20 countries with different national laws and policies about business expense and all of those. And then you overlay corporate policies and the whole thing just, and then your

business is just codified that way. And you can't replace it. Back in the, in the late 1990s, who now is sort of the godfather of enterprise software, Larry Ellison at Oracle, back then they were only a database company and entering the world of NetSuite and ERP and all this. But he went on a rant, a multi-year rant about how enterprise software

was so stupid because everybody customized it. And he had this saying that just said, businesses should just stick with the 80% solution and they should just use whatever works like 80% of the time. And most enterprise people were like, well, A, you're just talking your book because your software only does 80% of what I need.

But B, like, that's just not how it works. Like, if you take the auto industry and you just take the top 10 companies in autos, they all, you know, putting aside EV versus gas or whatever, they all just make cars, which is a lot of known technology with assembly lines and workers. Like, what differentiates

the companies, and what differentiates them is how they operate and the internal processes to decide what car to make, how many more materials to buy, what currencies to hedge, how many people to hire, when to introduce a new product line. All of that is enterprise resource management

and planning, and how is all that done? It's all in SAP. So those companies are effectively run by people sitting in conference rooms looking at SAP screens. and the difference between Ford and Toyota and General Motors and Daimler

are not just that they're looking at the same screen. It's that they chose which screens to look at, which customizations to make in those screens, and then they go and they buy steel and aluminum and wire and dashboards and radios from all the same places. And so I just think that people wildly underestimate the level of housing, of sophistication that customers apply to this software.

Back when we were first starting to get Excel used in companies, you'll laugh at this. We used to do these little visits and we'd go visit bankers. And so we're sitting in Goldman Sachs, and we're telling you about Excel is better than Lotus 1-2-3, blah, blah, that's super old. You don't even know it. I know what that's good.

Trust me, it's old. And the guy at Goldman looked at us and said, I don't think you understand. We make more money from Excel than you do. And we're just sitting there like, what is he talking about? Like it made no sense to us. And then we start to think about it. And it's like, well, we sell Excel to Morgan Stanley and JPEC and

Chase and everybody else. And the Goldman was saying is their application of Excel is so differentiated. And that's, and that wasn't just people typing. They built add-ins. They wrote all this code. They defined their work processes. So there's this wild underestimation about like you could vibe code your way into enterprise software. I was at a dinner last night and there was someone there who was like the head

of revops at a, I don't know, maybe growth stage startup. And his task, this is like a thousand plus person company, was to rebuild their Salesforce instance internally. And, you know, I think he's like, oh, well, you know, we know all the fields. We can import all the data. And I was like, that's not really the part that's tough, right? It's, well, how are you deciding what, like, what get it's captured, how the whole organizational lines around it. And then who's going to maintain this also over time? I think that's like a piece that just falls off. You can vibe code a CRM.

We've all vibe coded projects that have already gotten stale and we haven't touched again because it's painful. And it takes time and needs to adapt to the business. And Simo, you've also written about how, you know, there's an entire ecosystem of startups now that are just building on top of SAP and that are, you know, like sort of building around all of the kind of headache-inducing stuff, but still using SAP.

So, like, to both of your points earlier, like, these, like, legacy SaaS systems are just, like, so deeply embedded that the newer insurgents are just coming and they're, you know, building on top and around them rather than kind of, like, trying to rip out and get people to migrate completely. A lot of what we're seeing AI being used today for is, is how do you make it, like, I think the word is often used, like, conversational, or, like, how do you pull the information out and actually make it more usable? How do you retrieve the information from SAP without needing to go, you know, run a SQL query and get all the

information or, like, look at a bunch of screens? It's like, okay, well, if I want to be able to connect to Stevens point or analyze, like, you know, three different sets of tables and different geographies, like, can I quickly, you know, query that, like, a natural language way? Can I get reports automatically generated that are customized to me without needing to, like, go back through the, like, SAP customization process. And I think that is, that usability layer is, I think, indicative of what's happening now with software in general, which is accessing the UI is optional. And, like, going back to the SlackBots point, like, you want the information delivered to you versus needing to go to the UI. But the data and the business logic inside, either it's SAP or something else

that's being replacing it, that's building it, but that all still needs to exist one way or another. That's an incredibly important point for folks to take away, which is that the biggest thing about enterprise software is it almost always does what somebody wants it to do. They just don't know how to make it do that. Like there's no report that SAP can't generate,

no graph, no chart, no analysis, or whatever. but you just can't figure it out. Or maybe it's configured so you don't have permissions or something. So the way to think of it is in enterprise software, the two most frequently used features exist in no enterprise software natively. It's export to Excel and export as CSV or PDF, you pick. And so all enterprise software, the first thing they have to do that's missing

when they show up and they do that first demo is the customer say, does it do export to Excel, or does it do export to CSV or PDF? Because then you know you have an escape valve to do the thing that you couldn't do before on analysis. And so what's so cool about where we are today is that now with the language models, you have this incredible way to actually consume those

in much easier than you could before. If you think about PDF, like the old way used to be like, okay, I want to figure out, like, exception handling, some report that my system emits, you know, declined expense reports or whatever. But I want to do it over some weird time period or across different currencies that it doesn't handle or some weirdness that you can't figure out the UI.

So now you can export them all and take this 20 PDFs and put them in a model and do a bunch of analysis that you couldn't do before. Or if you did, it was all copy and paste and this mundane thing. And to something that Sima wrote about, these sort of, I don't remember the word you use,

but these ad hoc business processes are the ones that really become the most interesting because they're interesting because that's how a business runs, but they're also interesting because those are the next products. Like those are the next companies that people start. You know, CRM used to just be a spreadsheet.

Like, if you were in a business and you were account manager and you kept track of your accounts, you just kept track of it in Excel. And then a company got started to do that. It wasn't Salesforce first. It was the predecessor called Siebel. And then people like,

oh, we should make a whole company that does this. And that's what some of these apps are that you're seeing using language models and interfaces that are chat to SAP or to Salesforce. They are just trying to take advantage of what the alarms are really good at, which is synthesizing, and orchestrating unstructured information.

Yeah. I think we also forget that, like, Salesforce is, like, really an enforcement mechanism for, like, the go-to-market team, which is, like, okay, are you collecting all of the information that you need to? And, of course, we can talk about Salesforce hygiene, et cetera, as a separate point. But, like, do you have all, you know, is the human doing the, getting the data to then have, like, to, to then, you know, capture the state of the business. Mm-hmm. Which, okay, but I think if we like now switch to the agentic world,

and I think, again, we can talk about what agent means, but imagine there's an agent that needs to do outbound calling or outbound messaging. They want to be able to retrieve that information. They don't really care about how the fields are organized or, you know, how many clicks it takes. But they do want to be, they still need to access that information. But then the second piece they need is this context thing. So we've talked a lot. I feel like the Internet has talked a lot about.

context graphs over the last six months. But what is that? That's all like the exceptions. What do you do? How do you handle certain cases? It's the edge cases and the permissioning and all that stuff that needs to be and all the policies that are not necessarily in the fields of Salesforce. And so for the agent to then go, go back to this 80-20 thing, the agent can, you know, extract all the information, send an outbound email based on the information that's in the CRM around the person and their persona and what they do and all that. But then like, okay, how do you deal with a case, one case versus another and how they respond? And it's like, oh, well, normally if it's a person who's in Asia, we respond this way, but if it's a person in the U.S., we respond this other way,

that's not captured in Salesforce, but that's, that was in someone's head. And so that's the context that I think is really important now for agents to be able to act on behalf of this data. Oh, that's who, I mean, for Salesforce in particular, that's incredibly important. because I've never met a salesperson, an account manager, an account executive, who thinks that the default is the right answer for anything with their account. And even if they get the Japanese language right,

you know, oh, it's spring and the birds are chirping, but you're overdue on your payment, your license account is wrong. Even if you do that correctly, the rep is going to want to handle it in their specific way. And I think that this notion of exception handling is just the root of the challenge with agents, which is almost everything interesting in an enterprise is an exception.

Yes, yep. Like all the people are about exception handling. You know, it's basically like spend 15 minutes of McDonald's and watch people start in the kiosk and give up and then go watch what they really want. And they're like, well, I wanted McFlurry, but I wanted two flavors and mix them together. and that's not in the,

and it's always the exceptions. And everything about automation and enterprise is handling exceptions. It just is. It's the strangest thing. But enterprise pricing is a great example. Like, how much is it per se? Well, you have to call us.

Well, you call, then you talk, and then it's still an exception. Yeah. And that's exactly right. So these exceptions are not captured anywhere right now. Now I think if you say, if there's a voice agent, that is doing, let's say, compliance check calls for freight

as one of our portfolio companies does, they are now collecting the exceptions through their voice agent and getting some of that context. Or if they're looking, we can talk about computer-using agents, if you're observing humans and how they are clicking through software, how they are responding to things, and now we have this ability now to, you know, not only record data or, like, record interactions,

but then process it via LMs, then you're able to start collecting some of this context. But it's a lot, there, I mean, as Stephen was saying, it's ever, all the interesting work is around the exception. So it's not like, okay, you know, three days later, we've got it all. We've got all the context because also like sales cycles take a long time. Each exception isn't like handled with the frequency that you get the data immediately, right? And so you have to feel comfortable.

You have to get to that point where you're like, okay, we've observed enough interactions to actually capture, to understand the exceptions. And then on the sales side, the buyer trusts that, uh, that, that this piece of software can actually, you know, has captured all the context to handle it. Well, I don't just add some building on that, because I think it helps us to go back to this notion of headless and what are both the challenges and the opportunities. Because, of course, if you're an engineer,

which almost everyone talking about what's going on in the world in AI today is, you think headless and API, agent API, just interchange them. And so you think, oh, well, it's code. I can write the business process down. And the problem that you hit right at the beginning is that if you're not an engineer, you can't even explain the process that you use

to resolve a customer issue. And in fact, it's sort of very interesting to watch Amazon really do some of the best work on this because they really, really don't want to have humans. Like, you can't call Amazon for anything. Like, it's just hopeless. And so what they're doing is they're learning with everybody, like the best way to automate something

and it's their religion, it's their core principle, which is you just decide it in favor of the customer. You know, oh, they sent the wrong thing. So you go to the chatbot, you tell them, the chat bot understands that you got the wrong thing, and it just sends you a new one. And I think that that's so interesting compared to sort of old school exception handling.

And then they use the data to go and improve the internal shipping and handling and warehouse process maybe it's the product description, a zillion other things or reviews. And so I find that's what I find so interesting about the capabilities of AI is that it's driving a different definition and different behavior at companies

about how to handle exceptions. And I think when we get through sort of the 1.0 version of this, we're going to get to a new version where people are like comfortable letting AI do or decide things because they realize it's adding a level of predictability and repeatability to their enterprise.

It is, it's funny to think that maybe customer service gets worse in the short term because things stop getting default decided in favor of the customer. You know, like suddenly, suddenly you actually have to defend your case again instead of like you being, you know, reshipped via the sensitivity in toothpaste like you, you feel like you were owed. But I think, I think maybe then more generally, it sounds like you, you're both saying that automating the long tail is still kind of the hardest thing about, about all of this. Is that true? Or would you say that there are other hard things that developers

and founders also need to think about? I think that's part of it. I think there are a lot of other things around like permissioning. And this is all, I think you could probably lump it into the hard tail, but long tail, but like, permissioning is part of this, right? And like, you know, as you give people are given, you know, API access. It's like, okay, so which in which cases can people extract data, when can they write versus read? Like, that all needs to be figured out over time as well. And like interactions also between agents, right? And I think if you go back to the idea of a system of record, right, there's, it's one, ideally, one central repository of data that is the source of truth. Well, that now if you're, have multiple accessing and

writing to it, like who gets to access when? And I think that's a, anyways, these are additional problems that I think need to be solved. Solvable, but we'll take time. One of the things that happens in technology shifts is, you know, everybody knows the thing about nobody understands exponential when it's happening. So you have to be very careful to extrapolate and end up extrapolating linear when something

exponential is happening. But the same thing happens with productivity or an analogous thing happens with productivity, which is people look at the existing body. of work that happens today. And they say, okay, how do we make that easier? And then all of a sudden, there's all this fear that we're going to automate everything away, that everything is just going to become an API,

which developers and engineers say, oh, that will be easy. And then we'll be in this Nirvana world where everything is automated and easy and predictable. But they forget that productivity drives new scenarios. And so the minute that you can get something easier with automation, and you can actually automate it,

which I do think is happening right now with agents and with language models, well, then we're going to dream up a whole bunch of new stuff to do. Like I just mentioned this loop that Amazon must be in on customer servers. Well, they got rid of all the phone people and the phone experience that would be miserable to do a return and the challenge response and the fighting

and like, can I return this? And do I have to package it up or will you just ignore? Like toothpaste, they don't want it back. Like, that's a pioneering invention by Amazon is like, you know, if somebody gets the wrong consumable, we just don't want it. Like, they're poisoning it, they use part of it, it's cheaper to just have them throw it away.

Well, that never happened before. Like, you used to have to actually bring spoiled food to the supermarket and show it to them. And so they've fixed that level of productivity. But now there's this back end that's just out there constantly figuring out how to have it not happen again. And now they need a new, level of analysis, a new set of tools, and the long tail got no shorter. It just got longer in a

different way. Yes. And I think people forget that that's how innovation is this constant reinvention, and it's a growing pie, not a static pie. And all the negativity around AI comes from just thinking that the work to be done is this fixed thing that takes N people and M amount of software and we're just going to replace N people with M plus 5 and then we're done. There's no jobs anymore. There's just an agent running. And that's just never

going to happen. Like legal is a great example of this, like where people do contracts and they think that the law is going to help contracts get done quicker without lawyers. Except I can assure you contracts will get longer and more sophisticated and encompass

way more sets of scenarios than a person ever could. And there will be And that's going to create a whole. More litigation around it, and that creates a whole ecosystem. And more deals. Look, there's the now apocryphal, semi-apoccal famous example of radiology, which is a correlation, not a causation. But radiologists all love AI. And now we are having a radiology shortage.

It's not, there's a lot of reasons. It's complicated. But it just shows that the innovation wasn't static. And the market, for the demand. wasn't static. And so I think that a lot of what happens just in the micro at the enterprise level is, the minute you automate the most mundane thing

and think you have it all squared away, whole new things appear. Like actually, like expense reporting is a really good example. You know, first there's nothing, then people figure out how to like do spreadsheets, and then people figure out like, oh, now we have a whole system, we can analyze it. And now all of a sudden, business travel,

you get ahead of the curve. And you're like, well, now let's just use miles for business travel. Let's, you know, route our travel requests to the best prices we can get at any given moment rather than just default to one carrier. Let's use a specific credit card for business travel that buys us a bunch of different added benefits that we no matter to our patterns of travel. And so suddenly, like, there was a bigger job called business travel analysis that takes way more people than just booking the flights, which everybody can just do on their own. There's always another layer of analysis on top. Always, but the analysis then drives new

processes and new behaviors that themselves differentiate companies. Business travel, to stick with that example, is a huge sink in most companies. It's just a giant expense hole that they wish they could shrink. But once they can tie it to how things perform in their company, then it's more than just expense moderation, it's actually figuring out performance optimization. And figuring that whole thing out becomes like a different kind of job than just booking travel and analyzing expenses. It just becomes this whole remote work optimization tool. And then it's a different thing. I think the other thing interesting, not to double, you know, to spend too much time on business travel, but I think it also ties the physical and like digital worlds. Like there are always things that

there will be humans doing. You know, maybe it's not back office TPS reports, but like salespeople will be closing deal. There will be human interaction to close deals. People will be getting on planes as a result. And maybe they aren't spending as much time entering data into Salesforce or doing things along the way, but they will, there will be these humans doing online and offline work. And I think that actually is something that, you know, there will always be a data

exhaust from things to capture, optimization that needs to happen, and that isn't going away either. Well, I think open source software development is actually a really good example of this. Because the hardest thing in software development is, you know, you have to be finished at some point so that everybody knows this is a stable release

and can go build on it. And the art of finishing is this long tail of, like, not changing the code. And there's no API for that. like developers think they don't they developers don't hesitate to think there should be no API for that they they they can think of a way to automate it with voting and with a discussion that has sentiment analysis or whatever but they you still need a bunch of people to concur over a decision to fix or not fix something and yet they'll advocate those same people will just

say some other business process like closing the books for earnings that should just be an API and it's actually literally the same mental model. Like there's a bunch of stuff and we're deciding when to close the books and what sales to account for what and where. It's fixing a bug. And there's a story around it, a narrative, and we have to explain it to our boss

and if something goes wrong, we need a trail that explains who did what. And so so much of what a business really is are just the people deciding things. and all that software does is it up levels, abstracts, and changes what they decide and how and what tools they use. The other sort of follow-up to Seema's point is like it's the best case for just recording everything you do,

like just voice recording everything you do to like capture. If people are, you know, going and flying and closing deals in person, make sure the software or the LLM can kind of capture everything that happens at all times. Obviously, not advocating for, you know, full panopticon, but... But that is the context gathering. Exactly. Exactly. And whether it's like, you know, recording and, you know, conversations or taking

emails and, you know, written artifacts and ingesting them, this is all, that is the way the world is moving. Exactly. So, yeah. Well, it's also, to your earlier point, you know, expertise exists in this cloud in an organization. And it is the untapped. resource of the modern era. And Aaron Levy at Box has done the most eloquent job of explaining

repeatedly the assets that exist and all of these, you know, word and Excel documents streamed throughout a company. And it's actually very, very hard to understand which documents are important, which ones to believe. And part of being in a company and having a culture is really knowing the answer to that. And it's super interesting to watch the customers at Box, use Box to actually answer

those questions. You know, which are the sales PowerPoint presentations that are actually working? Which are the spreadsheets and the models that people actually rely on? And I think that AI is the first thing to come along that really taps

into that unstructured information in a company. I think before we wrap up, it might be good to visit the sort of more immediate history and then the more far away history of kind of what headless software even is. I know Stephen you wrote last year a piece in reaction to the rise of MCP servers, and in that piece you also related it actually to sort of like early micro-exam. Microsoft litigation that the Justice Department levied against them. And part of the argument was that Microsoft had a lot of products

that could be categorized as middleware. And I'm just kind of curious, you know, in all of these different software waves that you've witnessed, kind of in what ways is history rhyming and repeating? Maybe not on the litigation side, but on the, you know, product level. That part will continue to, yes. Yeah, it's super interesting. You know, and I love Semina Opine on where she sees things going with startups in this regard as well.

So I'll go quick. But the real thing with MCP is it's very much like everything we're seeing now is that so much of it is driven by an engineering view of what would make for a good software architecture. And very little of it is being driven by sort of the using Sima's the physical reality. of the world. And so of course, if you're an engineer and you would love to have, like, every tool you want to use, to have a very clean API, preferably like a command light interface

that pipes text in and out would be perfect. But that turns out to, like, not be how the world wants to work. There are many, many reasons why it doesn't want to work that way. Seema touched on many like security and compliance and things like that. But the reality is that no software, wants to be disintermediated by some other layer above it. Like, nobody wants to just be put in a corner

and said, your job is to just store this SQL format for expense reports and do nothing more. And then we're going to use you only for that. And then, by the way, we're piping you through to some other tool to analyze expense reports. Because that's not a growing business. That's a decaying business. And so, it's a...

This whole notion of like everybody is going to be perfectly content to be abstracted by some benign layer in the middle, it just doesn't really work that way. And it's because customers actually do not want to assemble their scenario from a bunch of different providers because all it takes is your system will only be as stable as the most unstable part of that. So if expense support company goes out of business,

you're completely out of luck. So you want your experience support company to be thriving and doing more stuff, even though you in your head, you're like, I wish they would just stop. I don't want any more from them. It's getting complicated, oh, they just did a UI reworking,

that's driving me crazy. And the flip side is, those companies, they're not just gonna sit there and decay. And they're gonna look to the left, and they're gonna look to the right, and they're just gonna do the stuff that they see people using with their product. And so SAP, the example,

Seema used. We're seeing this whole ecosystem grow up, and SAP is just going to do those things. And that's the net. Now, not all of them, and most of them, they're not going to do very well. In fact, just before this, I was talking to somebody and we reminded them that in most giant enterprise companies, they view just a tie with some competitor as a win, because they'll just bundle it into their existing thing and give it away. And so it's, but this middleware layer, it's always, always very unstable. It looks great in a network hierarchy diagram of the OSI levels of networking,

but it's just never that stable. Yeah, I think two things I'll add. So one is, yeah, the practical realities, like go back to the Salesforce example, or like Workday. Workday has had APIs that you could work with, but can you really actually extract all of the data out of Workday in a clean way and just operate without using Workday?

No. Workday makes it extremely difficult to actually, like, get access to the documentation and work with the, like, and they don't expose all the input. Everything about the API, to use the API example is, this is, you know, analogous to what we're seeing now, which is that it makes it a dumb database, right? And so they're not incentivized to do that. So I think what we're seeing is there's three paths in front of you. One, if you were a consumer or like a business that's looking at buy software. One is, okay, I take Salesforce and I either turn on agent force or build all my agents on top of it and then treat Salesforce as kind of the,

just the back end. I think to what we just talked about, some of that will work, but some of that will also not work because Salesforce doesn't want you to want that to be, you know, they don't want to be just the data in the background, right? And so, you know, I think that there will be mixed results around that, and I don't have, I'm not bullish on the income. competent software building great agents on top. There's option, too, which is you just totally DIY it. You have the most control in that situation. However, I think to everything we just talked

about, that's really hard, right? Like, you have to rebuilding true enterprise software. And I think for a startup building, rebuilding a CRM much easier for rebuilding, you know, a CRM for a Fortune 500 business. It's a lot of business logic to capture. And you're also trying to like do open heart surgery while like the patient is like alive, right? Or, you know, whatever you want to use the analogy. Well, hopefully they're alive. Yeah, yeah, yeah. Yes, yes, yes, of course.

But I mean, you're like taking the engine out mid-flight, whatever you want to say as the analogy. That's really hard. You have to get the like practical realities of permissioning and collaboration and all that right. Then there's a third option. I think this is why we are continued to do what we do in investing in AI software is because there is a reason that like, you know, agents can continue to be built.

the data can be sucked in and built in the background. A lot of what we're seeing right now is things that are working alongside at SAP or a layer of visibility on top that is enhancing the experience and allowing the business user to then run agents on top of the existing data they have but also not throw out all of the logic they've had in the background. And then I think also create a new system of record. Like voice agents are collecting new data, recordings are correcting, collecting new data, transcription, ingestion of documents, all of that documentation is pulling in. And maybe, you know, one day these AI startups will replace the systems of record in the back end, but they are doing so in like a systematic way of observing how the business is operating.

I guess to close this out, though, Seema, you sort of just touched on this. Where are we really seeing the biggest opportunities for startups right now? I think, look, a lot of this, yeah, what I was just saying, it's, it's doing the things that the incumbents are not doing right now, which is, which is going from a layer of collection of data and into how do we take action on top of it, right? And so take the CRM example, right? It's like, I'm not just logging all of the, like, call information, but then now I'm providing the intelligence background, okay, how do I prioritize leads? which accounts should we work on, what has a risk of churn, flagging all of that,

and then, like, sending the outbound, right? And so, and part of that is creating this agenic loop, which is you now, as the agent sends the outbound, sees the response, you're understanding, okay, A, what works, what didn't, what, how did people respond? And then, B, you're also collecting, like, benchmark data, too. I'm like, okay, this type of response is most effective in these cases, and in Asia, we should be using this language,

type of opening versus in Europe, et cetera, that sort of stuff, you're now agentially collecting all of that. And that's like an interesting data exhaust. So I think that's another area. And the third area I just would flag, too, is we talked about this like physical realities, but the other part of physical realities

is a lot of the vertical software that builds for the physical world, actually. And that is a really interesting set of data. That's not, it's like hard to capture, has been hard to capture historically. And, you know, you will have to, continue to pull together things that can be kept, you know, agents have been able to operate on software,

but then also what humans are doing out in the field, machines are doing out in the field, and pulling that back in. So like construction, manufacturing, all of that. Well, the universal truth for enterprise software is the most difficult thing to do that happens to be the dumbest is to attempt to just compete head on with an existing category. And by head on,

mean not just the same category, but doing it the same way. The biggest opportunity right now is always, always to look at the existing sort of mental map of enterprise categories and be in between to establish players. Because the thing that you know right now during a massive technology shift is the one thing that established players won't do is disturb their existing product line and go to market. So they absolutely will just be. just be bolting AI on top of their existing product. They won't be getting rid of it. They won't stop working on it.

They won't do anything to break it. They're just going to try to weather this technology storm by sort of power through it, powering through it. And so your opportunity in a startup is to just look at Tube League 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 the new way and the new way exclusively.

And by not attacking head on, you don't show up at every single customer and have them go, you know, well, you need to do these 8,000 things before you even enter the door. Instead, you have an equally difficult question, but when you're in control of it, which is why do you even exist?

And that, but you, that's your own question. You don't have to answer to a series of, 20-year frameworks that, a 20-year-old framework that got created to answer a bunch of questions that aren't even relevant anymore. And the best example of this is, is HTTP and HTML.

Client server existed, but the reason that those took over was not because it did all the things that client server did. In fact, it did none of them. But it implemented that concept in an entirely new way. And so the web exists in spite of the fact that, Legacy vendors had a trillion dollars invested on how client servers should work.

Well, and I would say the other piece, too, it's not just between two legacy vendors, but I think now there's like a layer of translation between two different functions within an organization, too. Oh, yeah, yeah, for sure. Software has always sold like, oh, I'm selling into just, you know, the sales team or the finance team, but then there's like these handoffs and which is now the context, right, but on bills and deals, and like that actually also presents an interesting opportunity. So the last question I have, which is for Stephen, is so network effects is this thing we always talk about on the consumer side. And it's a great, you know, source of defensibility.

No enterprise software business, as far as I can tell, is successfully, you know, implemented network effects. But you could argue that is a good source of durability over time, right? And I think Salesforce has tried this in a couple ways in the past. But do you think that like enterprise software will start entering the, you know, the field of network effects in terms of like, okay, we're going to have both buyers and sellers on our CRM and therefore be able to like mediate these transactions or like, yeah, I'm curious to get your take on that? Well, certainly network effects outside of a company are extremely difficult for a bunch of compliance and security reasons. But the biggest network effect in enterprise software is inside of a company. And we're seeing that happen now with just chat. Like all of a sudden you're seeing

it's so incredible to be at this dynamic that almost felt like the good old days when some very motivated person, like most people who work in enterprises, it turns out, are not super interested in making their job better. They actually just want to go to work, get paid, and go home. And they don't come to work every day going, ooh, how can I make my,

how can I streamline my task? They just want to not mess it up. That is a lot of the world. But there was a small set of people like those bankers at Goldman Sachs that were like, how do I do more deals, faster, better, more clever models. And so they were using Excel

when the other bankers were using one, two, three. There's actually flirting around on the internet is this old commercial for Excel, this launched TV ad from the late 1990s where, sorry, the late 1980s, where the first Excel spreadsheets were being, used. And it's a person sitting there with this monstrous laptop that weighed like 12 pounds in an

elevator trying to use it. I'm laughing because, of course, they were trying to not run a battery life in the elevator ride, which was invariably the case. But all of a sudden, this crowded elevator of a bunch of people in these 1980s ties and 1980s wearing glasses looking at the spreadsheet going, what are you doing? How are you doing that? And getting all excited. And fast forward to 2025, and that's exactly what happened with chat. In fact, I had a friend at SAP that was writing like a SAP white paper about something. And I just asked them,

tell me what questions you're trying to answer. And I did the prompts and sent them back a white paper. And I'm positive I kicked off some sort of viral loop, not technically a viral loop, but some sort of network effect viral loop inside of her team. Because all of a sudden, people are seeing how to make their job better, and it's accessible to them and they're doing it. So I think, and to your point, FEMA, like this idea of a tool that enables two functions

to talk together that couldn't before is golden. Like, that's exactly, like, that's literally what enterprise software integration is, except that's all manual brute floors, higher accenture kind of stuff. And so if you have products that bridge this, and, you know, Figma did a bunch of this with design and product development. And so if you can develop software that leverages AI in order to bring together parts of an organization that don't normally communicate, that's a new category. And we've seen that with things like IT budgeting, where IT and finance would end up with

tools that ended up helping them both do forecasting. And the cloud enabled that. And so I think that that's a huge opportunity. Well, I think that's also an amazing note to end on. Thank you so much, Stephen, for joining us here. Okay, Seema. Yeah, and thank you. And thank you, Seema.

Thanks for listening to this episode of the A16Z podcast. If you like this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcast, and Spotify. by follow us on X at A16Z and subscribe to our substack at A16Z.com. Thanks again for listening and I'll see you in the next episode. As a reminder, the content here is for informational purposes only. Should not be taken as legal business, tax, or investment advice, or be used to evaluate

any investment or security and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see A16Z.com forward slash disclosures.