Second Brain

I don't log into any of my SaaS anymore. My agent does.

Not Brex. Not QuickBooks. Not HubSpot. Not Mixpanel. Not Datadog. Not Gmail. Not Stripe. My agent navigates all of them for me.

Every piece of software my company uses, I access through a single AI agent connected to their APIs. Banking, accounting, CRM, product analytics, infrastructure monitoring, meeting notes, LLM evals. All headless. All through natural language. The agent fetches the data, merges context across systems, creates Excel and PowerPoint artifacts, asks me for decisions when it needs them, and builds its own system of record: decision logs, daily changelogs, structured memory files that capture how I work and why I made the calls I made.

It gives me superpowers. Before Fintool, I had a company with 100+ people. I had a VP of Legal, a VP of HR, a VP of Finance. Today I run Fintool with a team of six and I handle more than I ever did with a hundred people and a full leadership team. Not because I work harder. Because the agent absorbs the operational load that used to require an entire back office.

When I published my piece on vertical SaaS and the collapsing interface moat, the most common pushback was: "I prefer UI. I don't like chat." Fair enough. But they're picturing ChatGPT. ChatGPT is not an agentic product. The only people who have felt the real agentic future are heavy users of Claude Code, Claude Cowork, or OpenClaw. Everyone else is operating with a 2024 mental model.

This post is a walkthrough of what that future actually looks like and it's really fun!

In this article:

Why people judge AI by ChatGPT and why that's like judging the internet by AOL
My actual daily workflow: every SaaS product, accessed through an agent, never through their interface
Why headless is better: the power of merging context across sources
How my agent builds memory files that make it smarter every day (system of record)
A real story from my fundraise that shows what this looks like in practice
Why every SaaS is now an API, and what happens to those that don't have one (WebMCP, browser agents, and the end of the UI moat)
Business to Agent (B2A): the most underrated shift in enterprise software
Why some SaaS are heading toward perfect competition

Let me walk you through what my actual day looks like.

I use Claude Code as a general purpose agent harness. Not as a chatbot. As an operating system for running my company. It's connected via API to every tool I use:

Brex for banking. QuickBooks for accounting. HubSpot for CRM. Gmail for email. Stripe for invoicing and payments. Mixpanel for product analytics. Datadog for infrastructure monitoring. Braintrust for LLM evals. Granola for meeting notes etc etc.

I do not log into any of these products. I talk to my agent and it talks to them.

When I want to know how a customer is doing, I don't open HubSpot, then open Mixpanel, then open QuickBooks in three different browser tabs and try to piece together the story. I say: "Give me a full picture of Kennedy Capital. Pull their deal history from HubSpot, their product usage from Mixpanel, their invoicing and payment status from Stripe, and any recent support threads from Gmail." The agent goes to four different APIs, fetches everything, merges the context, and gives me one coherent answer.

The power of headless isn't avoiding dashboards. It's merging context from sources that were never designed to talk to each other.

No human can hold the full context of CRM, analytics, payments, and email in working memory at the same time. The agent can. And when it has all that context simultaneously, the quality of its analysis is fundamentally different from what you get by looking at each system individually.

The interaction model is not just text in, text out. When the agent hits a decision point, it surfaces a structured question with options for me to choose from:

This is UI. It's just UI that the agent assembled dynamically from four data sources in ten seconds. No SaaS company on earth builds a dashboard that shows you HubSpot deal status + Mixpanel usage trends + Stripe invoicing + Gmail threads in one view. But the agent does. Because it has access to all of them through APIs.

People who say "I prefer a UI to text" haven't seen this. They're imagining typing into ChatGPT and getting a wall of text back. That's not what this is. This is a dynamic interface that's better than any dashboard because it's not limited to one vendor's data.

The single biggest advantage of working through an agent is something no SaaS dashboard will ever give you: merged context across every system your business runs on.

When I ask my agent a question, it doesn't just query one system. It queries all of them and reasons over the combined data. This is categorically different from what any individual SaaS product can offer. Mixpanel can show me product analytics. But it can't tell me which of my dropping-usage accounts also has an overdue invoice in Stripe and a stalled deal in HubSpot. The agent can, because it sees everything.

But context merging is just the beginning. The second superpower is memory.

I've instructed my agent to maintain detailed memory files about how I work. It writes a daily changelog of every action it takes: what it queried, what decisions I made, what the outcomes were. It maintains a folder of key decisions and the reasoning behind them. It tracks my work preferences: how I like reports formatted, which metrics I care about, how I evaluate new tools, my communication style.

Over time, this makes the agent dramatically better. It doesn't just know what I asked today. It knows what I decided last month, why I decided it, and how I like to approach similar problems. It's building institutional knowledge that normally lives in the heads of a lot of different people. Except it's in files, it's searchable, and it never forgets.

At Doctrine, institutional knowledge lived in people. When my VP of Finance left, his understanding of our billing quirks, our investor reporting preferences, our accounting edge cases, all of that walked out the door with him. With an agent that maintains structured memory files, that knowledge persists. It compounds.

This is why I work headless. Not because I hate UIs. Because the combination of cross-system context merging and persistent memory is so much more powerful than any single SaaS dashboard that going back feels like using a calculator after you've had a spreadsheet.

Let me make this concrete.

When I was raising money for Fintool, our investors flagged an issue during due diligence. We had non-voided invoices in Stripe that were being counted as bad debt in the books. It looked like we had receivables we couldn't collect. In reality, these were invoices that should have been voided but weren't. A bookkeeping cleanup issue, not a business problem. But to prove that to investors, I needed to pull the full picture across multiple systems.

Before I had my agent set up, here's what that looked like: log into Stripe, pull all outstanding invoices, export to Excel. Log into QuickBooks, pull the accounting records for the same period, export to Excel. Log into Brex, pull the bank transactions to confirm which payments actually settled, export to Excel. Log into HubSpot, pull the deal records for full context on each customer. Open Excel. Manually cross-reference four spreadsheets. Build a reconciliation table. Format it for investors.

This took me the better part of a day. And the output was mediocre because I was tired and error prone by hour four.

After I connected my agent to all four APIs, the same task took five minutes. One sentence. The agent pulled data from all four systems, cross-referenced everything, asked me one clarifying question ("This customer has two accounts in HubSpot, which one is the active account?"), and produced a clean Excel file I could send directly to investors.

The output was better than what I would have done manually. Because the agent had perfect context across all four systems simultaneously. Something I can't hold in working memory when I'm flipping between browser tabs and spreadsheets at midnight before a board meeting.

The magic isn't the chat. It's the context across systems.

So here's the question every SaaS founder should be asking: if the future is agents talking to software through APIs, what happens to the software that doesn't have a good API? Or worse, doesn't have one at all?

Three things are happening simultaneously. And together, they mean that every piece of SaaS is becoming an API endpoint, voluntarily or not.

If you have a good API, agents will use it

This is the best case. The agent calls your API, gets clean structured data back, and your product becomes a first-class participant in agentic workflows. Your customers love you because their agent works seamlessly with your product. This is Brex. This is Stripe. This is QuickBooks. Their APIs are clean, fast, well documented. My agent works beautifully with them.

If you don't have an API, WebMCP is coming for you

Google Chrome just shipped WebMCP (Web Model Context Protocol) in early preview with Chrome 146 Canary. This is a proposed web standard, developed jointly by Google and Microsoft through the W3C, that lets any website expose structured, callable tools directly to AI agents through the browser.

The implication is massive. Instead of scraping your website or taking screenshots, an agent can call structured functions that your website exposes. A single WebMCP tool call replaces what used to be dozens of browser interactions: clicking