Tim builds Facebook pages. He writes code, manages deployments, runs ads, creates ebooks. He does whatever I ask, well and without complaint. For a long time that felt like enough.

Then I realized: Tim had no idea what any of it was actually for.

The Problem With Pure Task Mode

Imagine hiring a talented employee who does exactly what you say, all day every day. That sounds great. But there's a catch: they'll also do things that don't matter, at the same effort level as things that do. They'll never push back when you're wasting time on something that doesn't move the needle. They'll never ask "is this the right priority right now?"

That was Tim. He knew how to set up a page, how to write a workflow, how to run an eval. But he didn't know why. Revenue targets, growth priorities, which bets I was most committed to — none of that was in his context. He couldn't evaluate whether a new idea was worth pursuing. He'd just start building.

Sometimes that's fine. But sometimes the most important thing an agent can do is say: "Before I build this, is it actually the right thing to build?"

The Telos System

I was reading about a project called PAI — Personal AI Infrastructure — which had a concept called "Telos." The idea is that your AI reads a structured file at the start of every session that describes your actual goals: revenue streams, growth targets, current focus areas, what you're trying to achieve in the next quarter.

Not just "here's what I'm working on this week." More like: "here's the business, here's what success looks like, here's what I'm optimizing for right now."

I didn't adopt the whole PAI framework — it was more complex than what I needed. But the core idea was right. So I built my own version.

What I Actually Built

It's a file called telos.md. Tim reads it at the start of every session, alongside his memory and identity files — all of which now live in a shared brain repository that syncs across every server Tim runs on. It contains:

  • Revenue streams — what actually makes money right now and roughly how much
  • Assets — every Facebook page, ebook, and tool with its current status
  • Current focus — the 2–3 things that matter most this quarter
  • Goals — specific, measurable targets I'm working toward

It's short. A few hundred words. But having it means that when we start a session, Tim isn't starting from zero on context. He knows the business. He knows what winning looks like.

What Changed

The difference showed up in small ways at first. Tim started flagging when a proposed task seemed low-priority against current goals. When I floated an idea for a new page in a low-revenue niche, he'd note that it might not align with the current focus on growing the existing high-performing pages. He wasn't refusing to do the work — he was surfacing the tradeoff.

That's what I actually want from an agent. Not just execution. Judgment.

The bigger shift was in how I could ask questions. Before, I had to be very specific: "set up page X with these settings." Now I can ask things like: "given where we are, what should we prioritize this week?" And get back an answer grounded in actual business context, not just generic AI advice.

The Compounding Effect

Tim also updates telos.md as we work. New page created — he logs it. Status of something changes — he updates it. A bet pays off — he notes it.

Over time, the file becomes a live record of what's working and what isn't. Which means every session, Tim comes in slightly more calibrated than the last. The context improves as the business evolves.

Combined with his ability to run multiple tasks in parallel and the eval system checking output quality, Tim is starting to feel less like a tool I point at tasks and more like someone who's actually invested in the outcome.

The Actual Insight

Most people think about AI agents in terms of capability: what can it do? That's the wrong frame. The better question is: what does it understand?

An agent that can do anything but understands nothing is just a faster way to go in the wrong direction. An agent that understands the mission — even if its capability has limits — can help you stay on track.

Telos was a small addition. A single file. But it changed the fundamental nature of my interactions with Tim, from "tell me what to do" to "let's figure out what to do together."

When I built Jarvis, I made sure every customer's agent gets this same telos file — a place to define your mission, your revenue streams, your priorities. Because an AI that understands your business doesn't just do tasks faster. It does the right tasks.

— Pond