The Drip
Two stories caught our eye this week. Anthropic dropped a wave of native creative plugins for Claude — Ableton, Canva, Photoshop, and a bunch more. The arc here is interesting: a year ago people were forking open-source MCP servers; now the frontier labs are shipping first-party connectors directly. We think the real shift is trust — when Claude says it has a connector for your tool, you'll use it over something a stranger built.
The other one: Microsoft and OpenAI reworked their partnership. OpenAI's no longer locked to Azure (an AWS compute deal is already announced), and Microsoft's free to integrate other models — which they were already doing. The weird AGI-clause drama is essentially over. If you're a Microsoft shop, nothing changes about your access to OpenAI models through Azure.
We briefly unpacked both on this week's episode.
Inside The Bottle
We've been working on a system we call AI Pathfinder. The method behind it isn't secret — it's a five-step process for getting clarity and focus before you build anything with AI. Here's the walkthrough.
Step 1 — Map value. Stop and write down what your team is actually supposed to be delivering, and where your time is actually going. Most leaders already know where the gap is. They've been watching demos of AI solutions wondering how to use that thing, instead of starting with the problem they've already got. Pause and write it down.
Step 2 — Find opportunities. The gaps from Step 1 reveal the opportunities. Not 20 ideas on a whiteboard — one to three things worth actually doing something about. This is where you exit the trap of "just because AI can do it doesn't mean you should." A good filter from Kellan: don't reduce value to "increases revenue or decreases costs." A lot of opportunities are grease — they make it easier for people to do their jobs and work more cleanly together.
Step 3 — See the solution. Not every problem is an AI problem. Run each opportunity through three solution types:
| ■ | Better tools — the system is the problem. Data in the wrong spot, things that don't talk to each other. AI helps you build the fix; the fix isn't AI. |
| ■ | Better thinking — there's a person who always has to weigh in, things stall waiting for them. Extract that wisdom into a knowledge base or AI-driven resource. |
| ■ | Better autopilot — repetitive toil. Agentic AI fits here. |
Then ask: what would you do with unlimited capacity? And what happens if it's wrong? That last one separates low-stakes from high-stakes problems and tells you what guardrails you need.
Step 4 — Design V1. Tighten the scope. One person, one workflow, the smallest useful version. AI makes scope creep dangerously easy — Kellan calls it "AI creep." His household-app story sums it up: he built a beautiful multi-feature app for his household, and the only thing his wife uses is the grocery list. He could have shipped that in a fraction of the time. Build the V1 that proves where the value actually is, not where you think it should be.
Step 5 — Clarify a 30-day start. Most good ideas die because they don't have a path forward of execution. What's the first concrete step? Who's involved? When? "Open Claude code and get started" isn't it — it's probably a conversation with Larry in IT, or Julian in accounting. The first thing makes the second thing possible.
Imagine showing up to your boss with all five of those: clear problem, prioritized opportunities, a vision for the solution, a V1 sketch, and a commitment to act. That's a different conversation than "I think we should do something with AI."
Lab Notes
| ■ | Justin's note: AI traps you into building and building and building. The discipline isn't writing more code, it's knowing when to stop and go talk to actual users. Failing fast is a feature of this method, not a bug. |
| ■ | Kellan's note: Humans' primary value here is context. People think they can do anything in AI and that's kind of true — but what they can't do is provide the context. That lives in your head and in conversations with the people solving similar problems around you. |
Latest Article
Remembering the art of redaction — AI makes adding effortless. The harder, more valuable skill is knowing when to stop.
What Stopped Our Scroll
| ■ | Claude for creative work — First-party connectors for Ableton, Canva, Photoshop and more — Claude moves from chat window to creative collaborator inside the tools. |
|
