I Built Jarvis (Sort Of) — How AI Runs My Two Businesses

You know the scene in Iron Man where Tony Stark walks into the lab and just starts talking? "Jarvis, pull up the schematics. What's the status on the Mark III? Run the numbers on the power output." And Jarvis — who knows everything about Tony's work, his history, his preferences — just handles it.

I built that. Minus the flying suit.

It's not a $50 million AI lab. It runs the operational backbone of both my businesses every single day. And it's the same kind of system I build for clients.

What it looks like in practice

I run two businesses. A SaaS product with 60+ customers and a consulting practice. Both need daily attention. Neither can slip.

Monday morning, 11:30. I have a call coming up with a prospect about a potential webinar partnership. I open my AI system and say: "I've got this call with Matthew. Pull up my notes from our previous conversation and give me some webinar ideas."

It already has context from our prior call. It generates five webinar concepts in about 30 seconds. Two are genuinely strong. I say: "Expand on three and four, give me a five-slide outline with key talking points." Done.

The whole interaction took three minutes. If I'd done that manually — digging through notes, brainstorming, drafting slides — that's an hour of work. Compressed into three minutes, and the output was arguably better because the AI had perfect recall of the prior conversation.

But here's the thing that really got me. It wasn't just the speed. It was the voice.

The three-component unlock

There's a specific combination that makes this work. It's not just AI. It's three things working together:

1. Voice interface. I talk to this thing. I'm on the train, I finish a call with an investor, and I just start talking. "Add two to-dos: follow up with Peter about the deck by Thursday, and send him the case study." It creates the tasks, files them under the right project, and I never open an app.

Research from Stanford shows that speaking is 3-4x faster than typing. That's not a small difference — that's the difference between getting something done in the moment and adding it to a list you'll forget to check.

On Mac, Monologue gives you system-wide dictation powered by Whisper (the same speech recognition model behind most AI transcription). On Windows, Whisper Writer does the same thing. You press a hotkey, speak, and the text appears wherever your cursor is.

2. Headless tools — APIs and MCP servers. This is the part most people miss. For AI to actually do things in your business, your tools need to be accessible without a user interface. That means APIs (a way for software to talk to other software) or MCP servers (a newer protocol that lets AI tools plug directly into your business applications).

My bookkeeping software has an API. My task manager has an API. My metrics tool has an API. My email and calendar have APIs. That means the AI doesn't click buttons in a browser pretending to be me. It talks directly to the systems, the same way they'd talk to each other.

If your vendor's tool doesn't have an API, that's a problem. Not just for AI — it means your data is trapped behind a user interface that a human has to click through manually. Every. Single. Time. (I wrote more about this in The Future of Knowledge Work Is the Command Line.)

3. An AI tool that connects to everything. I use Claude Code — it can read files, run scripts, connect to APIs, and remember context between sessions. When I open a session, it reads my business files and knows: which prospects need follow-up this week, what the current metrics are, what I decided last month about the product pivot, and what my quarterly goals are. Other tools can work this way too — anything that can plug into MCP servers and maintain context will get you most of the way there.

The magic isn't any one of these three components. It's the combination. Voice means you move at the speed of thought. Headless tools mean AI can actually execute. A connected AI tool means it has the context to do the right thing without you spelling out every step.

You talk. Things get done.

What it actually handles

Sales pipeline. It tracks every prospect with full context — discovery call notes, emails sent, proposal status, follow-up dates. When I paste a prospect's email in, it stores the full text so future sessions have the exact words. It drafts follow-up emails. It maintains a schedule and surfaces what's due during morning standups.

Metrics. It pulls live data from my analytics tool — revenue, customer count, churn, lifetime value. I say "show me the last 12 months of MRR" and get a table with trend analysis.

Task management. Connected to my task manager. Creates, completes, and organizes tasks across both businesses. At the start of every session, it already knows what's on my plate.

Structured rituals. A daily standup, weekly review, monthly retrospective, and quarterly planning session. Not templates — interactive sessions where the AI asks questions, challenges my answers, and pushes back when my priorities don't align with my stated goals.

Content. It manages a pipeline from idea to published draft. One client interview transcript produces five different content assets.

Bookkeeping. Half a day of fighting the bookkeeping UI became 20 minutes of talking to AI. Same outcome. Fraction of the time.

Strategic thinking. Because it has all the context — both businesses, the financials, the pipeline, past decisions and why I made them — it pushes back with substance. Not generic advice. Specific pushback like "you said last month you were going to focus on lead gen, but you've spent this week on product features — what changed?"

This isn't a developer tool

Most people think AI systems like this require a technical team to build. They don't. They require someone who understands how to connect the pieces — voice, APIs, AI — and who knows your business well enough to set up the right context.

The hardest part isn't the technology. It's thinking clearly about how your business actually runs: what processes happen every day, what information lives where, what decisions get made and how. The act of setting up the system forces that clarity.

What this means for your business

I built this for myself. It increased my output 4x. Not because I work more hours — because the time between thinking something and having it done collapsed.

The same approach works for any business. The specifics change — your processes, your tools, your workflows — but the architecture is the same. Voice in. AI processes. APIs execute. Results out.

The companies that figure out how to operate this way are going to run at a fundamentally different speed than the ones still clicking through screens and copying data between tools. That's not a prediction. That's already happening.