How AI Agents and Open APIs create new rebundling opportunities
Much of the AI hype of 2023-24 has been centered around horizontal capabilities of foundational models.
The real opportunity of AI lies not in creating mediocre marketing copy, but in the ability to reconfigure value creation across vertical value chains.
While most AI hype is centred around horizontal B2C use cases, the real opportunity lies in vertical B2B AI.
This post unpacks the the opportunity of vertical AI and how companies can win this game.
And with that out of the way, let’s get started:
Short-term and long-term games in vertical AI
Most vertical AI players today focus on developing vertical advantage with a proprietary fine-tuned model and a targeted UX.
Smaller models, trained on domain-specific data deliver better performance on latency, accuracy, and cost than larger foundational models. This verticalization has its own reinforcing feedback effect. The more you develop vertical advantage, the more competitive you get on all parameters.
To deliver the most compelling vertical solution over time, the more the model is fine-tuned, the more deeply coupled future UX changes should be with the model in order to deliver the benefits of that model into the user workflow.
The advantage that vertical AI players have is that they operate full-stack – which is to say they provide a fully-integrated solution across the interface, proprietary models, and proprietary data. This creates the above flywheel of increasing defensibility as continued right to own the interface grants the tight to keep gathering proprietary data that helps further fine-tune the model.
This is how you win the short-term game in vertical AI.
But that doesn’t guarantee a long-term win.
Despite this flywheel, all vertical plays eventually end up participating in someone else’s ecosystem. This is a natural outcome of the abundance of vertical solutions that eventually proliferate. The end user looks to get their work done across one or two interfaces, instead of hopping across individual interfaces for every new use case.
Eventually, the only way to win in the long term with a vertical solution is to go horizontal.
Vertical strategies and AI-driven rebundling
To understand the opportunity in vertical AI, it’s helpful to look at the rise of vertical Saas over the past decade.
Vertical software repeatedly followed a playbook of
(1) gaining adoption with a focused use case, and
(2) rapidly rebundling adjacent capabilities around that initial use case.
Players like Square (with payments as a starting point), Toast (with POS software as a starting point) or ServiceTitan (with estimates as a starting points) have followed this playbook to scale out.
Eventually, all vertical plays seek to go horizontal.
The reason for that is as follows:
Most ‘disruption’ of the status quo happens through unbundling. But most venture returns are realized through rebundling.
Vertical plays unbundle. But in order to capture value at scale, you need to go horizontal and rebundle.
There is no sustainable value capture in unbundling. Unbundling unseats incumbents but doesn’t create scalable and defensible value pools.
That is achieved through rebundling. Rebundling involves bundling multiple unbundled capabilities into a cohesive customer-centric offering.
Most important, the successful ‘rebundlers’ establish a hub position and gain primacy of user relationship.
Venture capital chases unbundling because unbundlers hold the promise of rebundling and capturing value. Yet, most venture money is lost because a tiny handful rebundle.
This is the end game with Vertical AI as well.
You go vertical in order to go horizontal.
AI creates a new locus of rebundling
The opportunity for creating long-term competitive advantage in vertical AI is not so much in solving an initial pain point deeply and effectively. A lot of players will succeed in creating proprietary, fine-tuned models to solve for that. It’s a necessary requirement, but it’s not sufficient.
Long-term competitive moats in vertical AI are created by rebundling around the initial use case, using the unique rebundling benefits that AI provides.
AI creates a new locus for rebundling and the players that effectively leverage it around initial success in solving an early use case will be the ones that effectively dominate vertical AI.
Rebundling workflows in the age of AI
Business workflows are scattered across silo-ed software.
To actually get work done, these workflows have to be rebundled towards a business goal. You take the output of workflow X, Y, and Z, and make decisions or take actions towards a goal. This bundling of workflows towards a goal is performed by human managers.
Managers in organisations achieve this bundling by solving two problems:
The plumbing problem i.e. coordinating across silo-ed workflows
The goal-seeking problem i.e. using diverse workflows to achieve an organisational goal
In today’s B2B workflow, managerial resources in an organization serve as the locus of rebundling.
But technology is transforming the landscape of B2B workflows in two very important ways.
1. Increasing interoperability
First, workflow interoperability increases as we move towards increasing usage of API-based distribution of business capabilities.
Business capabilities are increasingly being opened out through APIs for external consumption.
An API provides an interface to a business process or capability, while also defining the contract for engaging at that interface.More here.
APIs are pipes through which capabilities can flow beyond enterprise boundaries. As these pipes multiply, they create immense opportunity for plumbing. Different pipes can be plumbed together, with innovative business logic, to create fundamentally new workflows and capability bundles.
We’ve seen two important changes to the rebundling of workflows as a result.
First, companies like Salesforce and Hubspot, which sit at a central position in the workflow enable integration of other players to create a seamlessly integrated workflow.
Second, companies like IFTTT and Zapier allow users to create if-then trigger-based workflows, combining across multiple workflows.
Both these solutions solve the plumbing problem of coordinating across diverse workflows.
2.The rise of AI agents
Increasing interoperability solves the plumbing problem. It doesn’t solve the goal-seeking problem.
This is where AI agents create a unique opportunity for vertical AI.
Agents are goal-seeking, and that’s what makes them different. While most technology aims at task substitution, agents go beyond tasks to seek goals.
Every agent operates with at least one goal (and possibly more than one).
In order to accomplish this goal, an agent must (1) scan the environment, (2) plan and deconstruct the goal into constituent tasks, and (3) act out the plan leveraging other agents and digital resources.
In an environment where companies are increasingly provisioning APIs to serve key resources and capabilities, an agent performs the three functions above as follows:
(1) Scan the environment: It scans its environment to identify resources (including third party APIs) available at its disposal.
(2) Plan and deconstruct: It plans and deconstructs the goal into constituent tasks and sequences and roll-ups in which they should be performed.
(3) Act out the plan: Leveraging the resources (including third party APIs) available at its disposal, the agent acts out the plan to achieve the goal.
In effect, agents rebundle workflows across APIs to achieve an end outcome. This is a powerful new vector of performance.
The most powerful outcomes of vertical AI will be observed in domains of high interoperability where sophisticated and well-trained agents operate across the open resources.
Agents create a new locus of rebundling.
They enable vertical AI players to go horizontal by coordinating across multiple workflows.
This is already happening. Have a look at this Twitter thread on Self-Learning Agent for Performing APIs (SLAPA), from March 2023:
Every goal is a bundle of tasks. And a managerial role – responsible for getting work done in an organization – is a bundle of goals.
When technology substitutes underlying tasks (red boxes below), the scope of the role remains largely unaffected as long as goal-seeking is critical to the performance of the role.
Let’s take the travel planning example. As new tools come in – travel booking tools, calendar management tools, payment tools etc, – specific tasks get simplified and even substituted by technology, but the goals within which these tasks sit are still managed by humans.
AI agents are different.
AI agents attack the goal. If an agent effectively takes over a goal, that goal no longer needs to be performed by the worker and no longer needs to be bundled into the human-managed role.
Effectively, a goal-seeking AI agent can unbundle a goal from the role.
The AI agent creates a new locus of rebundling.
There is fundamentally what “Sell work, not software” – a popular AI rallying cry – is all about.
Vertical AI players have a unique opportunity and that opportunity is not limited to merely developing a proprietary fine-tuned model. The real opportunity here lies in leveraging AI agents as a new locus of rebundling to integrate diverse workflows and coordinate work across them.