Most of Our Clients Have Good APIs
Most of our clients run on NetSuite, Odoo, Zoho, Dynamics 365, Salesforce. They all have well documented APIs, so we barely had to rely on computer use when automating workflows.
I also haven't targeted large enterprises with 10,000+ employees yet, so nobody has come to us running legacy platforms like Oracle EBS or SAP (both known for limited API integration). Such platforms also have little incentive to change this, as this helps maintain customer lock in and ongoing renewal revenue. Automating such workflows require computer use which is slower, more token expensive than API calls.
The Data Lake Idea
One idea I've been thinking about recently is using a centralized data lake as the primary source of truth for the business, with AI agents and applications layered on top of it to read, write, and reason over the same underlying data.
This data-lake-as-ledger realization is pretty recent for me, and I still need to flesh out my thinking.
At our company, everything goes directly into NoSQL databases, SQL databases, and storage buckets from day one. I just finished corporate tax and account reconciliation for our reporting period, and it was genuinely a breeze because everything was categorized perfectly from the start. That alone was the proof of concept for me.
I'm currently writing a guide for new hire onboarding, and the AI agents easily extracted industry-specific patterns from every client document because of how my databases were set up. I simply had the agents add an industry attribute to the relevant records without needing to change the schema. I didn't expect my earlier database decisions, especially using MongoDB for flexible document storage, to end up helping me this much later on.
What Operating Like This Feels Like
Whatever I'm building for myself, dashboards tailored to my exact needs, inputs and outputs, everything, feels completely different from operating inside traditional software. The business feels lean and fluid.
I've always hated middlemen, bureaucracy, and unnecessary layers of management, and this infrastructure style removes much of that structurally.
I also only run modern software with clean APIs anyway: Slack & Linear. If it doesn't have a good API, it doesn't make it into the stack.
Why Leadership Teams Are Terrified
On the enterprise question, I expect many leadership teams find this transition genuinely terrifying and quietly hope someone else will deal with it after they retire. Consolidating decades of proprietary data scattered across multiple ERP and CRM systems into a unified foundation would likely be a Herculean task, and becoming truly AI native could realistically take 5-8 years for many large enterprises.
But I don't see how enterprises running on legacy infrastructure compete long term with AI-native startups that treat the data lake as the ground floor.
The Cost Argument
The cost argument is real too. Add up decades of ERP licensing, support contracts, implementation consultants, middleware teams, and internal IT headcount just to keep old systems alive, and the migration cost starts looking very different. A lean system that replaces all of that could potentially increase operational capacity by 10x or even 100x once AI agents have real agency.
High-Agency AI Agents
That's where I eventually want to get: high-agency AI agents that query the data lake themselves, make decisions, and read and write whatever is appropriate.
At that point, dashboards almost become relics. They're built for humans who need to visualize data. An AI agent just queries the database directly. Why would it need a chart?
5-8 Years
For companies with 40 years of operations, that future is genuinely scary to imagine. That's exactly why it requires a high-conviction founder, board, and C-suite with the fearlessness and patience to commit to what is realistically a five to seven year infrastructure project.
Most organizations won't do it.
The ones that do may end up with a structural advantage that compounds for a very long time.
The Principle Matters, Not The Tool
I know my MongoDB instances will eventually start slowing down once I hit tens of millions of records or more. I'll move to Databricks or whatever best fits the scale at that point.
The specific tool matters less than the principle: own your data, structure it from day one, and let the infrastructure evolve as the company does.