Eric and Seema Amble's a16z essay on why the world still runs on SAP is a serious piece of work. SAP, Salesforce, and ServiceNow survive not because anyone enjoys using them, but because they still hold the records, rules, approvals, and process memory large companies actually run on.
Canonical data means the records a business ultimately treats as the official version of what happened.
They are also right that AI leverage shows up across the whole lifecycle, not just in day-to-day usage. Implementation, migration, maintenance, and extension all matter. The opportunity is not only net-new apps. It is making the systems companies already depend on more usable, more programmable, and less miserable.
The real problem is not just that SAP and its contemporaries are ugly or that there is too much clicking.
IBM just completed its $11 billion mega-acquisition of Confluent which reinforces the same thesis from a different angle. IBM's stated rationale was trusted communication and data flow between environments, applications, and APIs in real time. That is the connective tissue enterprise AI depends on. If you want a system to act proactively and still be trustworthy, it needs live data, governed streams, and clear lineage more than it needs another interface.
Until recently, we've been tacking sidebars onto the same old interface like copilot autocomplete, the Q&A search bar, then the agent that clicks around old software for you (brand new but already feels old).
One of our mantras at Big Robot is "the best UI is no UI". People should not be chained to their desks pushing pixels through a maze of forms and tabs. Data entry of any kind will be eliminated. We aren't even good at it compared to a machine. People should be communicating, refining, and strategizing.
The real win is when the software can pull the pieces together well enough that the person's job is no longer to operate the system. It is to define the outcomes and use the output.
The wedge is trust.
That fits with what I wrote in "Your Intelligence Budget." The human role keeps moving upward until it becomes judgment. In enterprise software, that means the machine should do the searching, pulling, and matching, and the person should review the result and decide whether it is correct.
Before a system can act on its own, people have to trust what it produces.
If a person needs to look at the machine's output and decide whether it is right, that sucks, but the system has to make that judgment possible. It has to show what it found, how the records connect, and why it reached that conclusion. Otherwise the reviewer is just being asked to bless a guess.
The goal is not another layer of software for humans to babysit. Let's skip adding ChatGPT to Excel and never touch a spreadsheet again.
That is the direction we care about too. Less software as a place you live inside all day, but rather a system that proactively brings you a recommendation, shows you why, and asks for judgment only when judgment is actually needed.