Comments on A16Z's Why the World Still Runs on SAP.
APIs are like the straits of Hormuz.
I read with interest A16Z’s take on SAP. See footnote for disclosure and my enterprise software related background.1
I’d urge you to read the A16Z post and then pop back here for my reflections.
Firstly, well done to the A16Z duo of Eric and Seema for writing this. I was pleasantly surprised that they had written it, and I found myself agreeing with much of what they had to say. I’m going to disagree with some bits as well. In parts, it damns SAP and other enterprise vendors with faint praise, and yet it assumes an incumbent continuity that is not assured.
This may mean I write more than one post. For instance, I want to do a post on upgrade and maintenance tooling with AI, but that can wait for another day.
In VC land we tend to get ahead of ourselves, and project an imagined future onto the present and believe it is already here, if not this afternoon, then first thing tomorrow.
There is indeed more longevity to enterprise software than most VCs will care to admit. Large chunks of government and banks still run on mainframes, and it is only relatively recently that cloud spend exceeded on-prem spend, I’d guess about 2021. A majority of SAP’s customer base still run on prem or hosted environments.
Here’s my take today: It is clear that there is strong likelihood that agents will take over a significant chunk of the interaction with ERP systems. They have not done so yet, but within a decade they will become a if not the major user of ERP systems. The use of Agents and other AI techniques creates a new layer in the application stack, the context layer. A16Z is spot on with this:
Enduring software in this category will look less like a chatbot and more like an operating layer: a unified data and action plane with a semantic model of business objects, plus guardrails that make AI trustworthy in production.
One of our founders, Jerome Gouvernel at datascalehr has written a thoughtful post about this layer.
How we interact with Enterprise software will change even more than did with the shift to web, but it has not shifted yet. Genuine agentic production deployments are still dwarfed by good old human users and traditional integrations still dominate API usage. But our world is changing.
Who designs, builds, controls, governs and captures the value in this layer is still not settled.
A16Z seem to assume SAP, Workday and others are unable or unwilling to respond to this shift. The vast capital that is flowing into startups, combined with massive changes in how one writes code, does strengthen the hand of start-ups to build and own that context layer, and capture a dominant chunk of those revenues, but it is not a slam dunk or a long hop (for cricket fans).
There is a non-zero chance that some of these incumbents reinvigorate themselves and build genuinely innovative solutions with AI at the centre. SAP managed the mainframe to client/server brilliantly, and the client/server to cloud transition adequately. In hindsight it would have been better off not having its finest engineering brains building a database, but to assume that it will just sit back and let start-ups shift it into irrelevance is folly. Similarly, I have spent my career under estimating Oracle, at my peril. Aneel returning to Workday may bring about more innovation. The ERP vendors will not go down without a fight. Their success will be largely determined by how quickly and effectively they can transform how they build software themselves.
On APIs and computer use agents, free riders?
Overnight I noticed that a bunch of founders and VCs have taken umbrage with what the CEO of Workday, Aneel Bhusri said about third party agents.
It is worth looking through some of the comments, but I’d suggest Garry and those that piled on need to read an ERP T&Cs over the weekend.
A month or so ago I wrote about SaaS and the impending doom.
The most innovative and effective department at the large incumbent vendors is their licensing / pricing department. If Startup x thinks they can sit a 3rd party agent on top of SAP or Oracle, switch off the seat licences, and gobble up that data and revenue, they are in for a nasty shock. I suspect those departments are building agentic licence auditors as we speak.
Today there are two ways agents can interact with an ERP system, either through an API, or by mimicking human user behaviour at a screen level, what A16z calls a computer-use agent (bit like RPA). If you call your agent Mary, the agentic employee, then don’t be surprised that it/she incurs some sort of user charge.
When agents blunder, the system of record will take the pain, so I have some sympathy for the ERP vendors wanting to control which agents come to the party.
I need to go back in history for a moment. Before the mid 1990s, enterprise software was used by a relatively small number of actual users. Managers and employees didn’t touch these systems, professional administrators did. The arrival of the browser as a UI created the opportunity for occasional users to engage with these systems directly. Initially, these solutions were built by third parties or even DIY. Some of these vendors accessed the ERP system via a superuser, or snuck in the database back door. It didn’t take long for the licensing teams at the ERP vendors to shut that tight, and create new type of user, the self-service user. The ERP vendors quickly annexed the employee and manager self-service user revenue. About a decade ago, SAP flexed its muscles with an aggressive licensing play called indirect access, it irked the customers, but most of them pay the toll.
SAP’s interpretation of indirect access:
1.8. “Use” means to activate the processing capabilities of the Software, load, execute, access, employ the Software, or display information resulting from such capabilities. Use may occur by way of an interface delivered with or as a part of the Software, a Customer or third-party interface, or another intermediary system.
The APIs in ERP systems are narrow, awkward and can be throttled, especially in cloud deployments. But some of the APIs are older than the Internet. Start-ups are facing a bit of a Straight of Hormuz moment.
It isn’t the same as 20 years ago, and the agentic opportunity is far bigger than the self-service opportunity, and some of the power dynamics have shifted. But there are strong echoes.
The open question today is how much power do the ERP vendors hold over the API and computer use solutions. There are limits to how much the ERP vendors can monetise this API and computer access, but it will not be for free. We have already see Salesforce flex here with slack, see Tom Tungunz’s post.
If the ERP vendors flex this muscle too aggressively, it will encourage customers to minimize ERP spend and usage, and eventually seek alternatives. The ERP vendors have leverage, but it is not unlimited. Many existing integrations use the APIs (the typical Workday installation will have more than 200 integrations, SAP many more). Policing APIs isn’t straightforward.
But if they don’t flex it at all, they are likely to lose the context layer to start-ups that will free-ride on the ERP stack. All the cool kids want to be context layer.
But there is a viable scenario where the ERP vendors step up a gear, and make a fight of owning the context layer. To do so, they will need to go shopping, and transform how they develop software, and get really smart on access monetization and control. It will not be easy, but it is feasible.
A bit of paradox
While underestimating ERP power in the short term, I think A16Z has over-estimated the defensibility of ERP incumbency in the longer term.
Castles were impregnable until the cannon, and then they weren’t. As we recently found out aircraft carriers were omnipotent until the submarine drone.
Here’s the other thought I can’t get out of my mind. As work changes fundamentally (assuming it does), the system of record simply can’t keep up. It becomes unfit for purpose. The weaknesses in APIs, limitations of computer use interaction, and data model rigidity mean that the ERP becomes more and more of roadblock. And someone really smart builds something so much better, and at the same time, AI solves the migration challenge, derisking leaving.
If you think about it for a moment, we built ERP to help us take the messy world of manufacturing business and store it in a structured form. We designed screens and processes to help humans translate business activity into fields. Business rules, validations and so on helped reduce the human capacity for error when capturing data, that’s why it’s so bloody complicated.
Now, with agents, we are asking a computer to pretend to be a human to interact with another computer, when humans aren’t very good at interacting with computers, and computers aren’t yet great at pretending to be humans.
There has to be a better way.
A couple of years ago I wrote on:
Systems of record and transaction in an AI agent world
Warning: Long post. Some metaphor overload. Thoughts in progress.
The more human-like AI becomes, the less like a computer database it will need to be. As we and AI learn, we and AI will fundamentally rethink what transaction processing and system of records need to be.
Wrapping this up
For now, the ERP vendors aren’t going away. A16Z are indeed right to note that the context layer will become vital, agents will do more of the system interaction, extensibility will become significantly easier, and system migration and support will be significantly improved with AI tools. I hold the ERP vendors have more fight in them than A16Z reckon, but a new wave of vendors are going to do very well on top of the existing systems of record.
But I’m also confident someone will come up with a 100x better way of doing systems of record (and the context layer) than we have today. I am hoping that we fund this vendor.
As usual I’ll end with a song. Had to be Dire Straits.
Disclosure, I spent a chunk of my career at SAP, and once upon a time I was Gartner’s lead analyst on SAP. I’m on the board of Basis Technologies, an established and highly successful SAP change management solution provider. I’ve stayed vaguely close to SAP since I left in 2018, but I’ve no formal engagement with the executive leadership these days. I have some insights into Workday and ServiceNow too. ServiceNow is an LP in our fund, and I covered Workday extensively while at Gartner, and I’ve maintained good contacts with the leadership over the years.
This was a golden opportunity for Dire straits' "Money for nothing", IMO :)
So insightful thank you. Love the passage “Now, with agents, we are asking a computer to pretend to be a human to interact with another computer, when humans aren’t very good at interacting with computers, and computers aren’t yet great at pretending to be humans.”