Why your AI keeps giving you mediocre answers

Hi everyone,

I'm rebuilding my entire AI stack from scratch, because I finally see what the right stack looks like for me.

More on that in a second.

First, the thing I want you to walk away with today.

Everyone is obsessed with prompts.

"What's the best prompt for X?"

"How do I write better prompts?"

Truth is, prompting is only part of the equation. And as the models get smarter, it matters less and less.

The real skill is context engineering: giving AI complete background information before you ever type a prompt.

Here's what I mean. You know markdown files? The plain text files with the .md extension? AI loves them. They're the simplest, most powerful way to give context about you, your company, and your work.

I have markdown files for everything. One describes my company, what we do, who we serve, how we talk. Another describes me. My background, my voice, my preferences. Another covers my sales process.

When I start a conversation with Claude, it reads those files first. It already knows who I am, what I care about, and how I work. The prompt becomes simple: "Draft a follow-up email for this deal." Claude knows the rest.

Here's the part most people miss. You don't just write context files. You tell AI when to pull which file, and why.

My sales agent reads the sales context file. My newsletter agent reads the brand guidelines and my voice file. My customer success agent reads the program context for that specific client.

This is why some people get great results from AI and others get generic slop.

Do this today:

You'll feel the difference immediately. Anything new you learn or build just plugs on top of that existing context. It stacks.

If you do this today, you're going to have a better time tomorrow.

The context engineering stuff connects to something bigger. I'm redesigning my entire system.

My Mac Mini used to run a setup called OpenClaw. Telegram bots, cron jobs, agents running overnight. It worked, but it was duct tape and scattered tools. Make for this. Zapier for that. n8n for something else. Mind Studio for another piece.

I'm consolidating everything into a clean, code-driven stack:

Mind Studio is out. Their latest pitch is "work with Claude Code to build in Mind Studio." Think about that for a second. If you need Claude Code to build in their platform, you don't need their platform. Just work with Claude Code directly. They also don't support custom domains, which is a dealbreaker for anything client-facing.

The goal is fewer tools, more control, and a foundation that builds on itself. Context files feed the agents. Agents run on trigger.dev. Data lives in Neon. Everything connects through code, not through 47 different OAuth integrations.

Speaking of tools. Notion announced pricing for their AI agents. I did the math on my setup. The SDR, AE, and CRO agents I talked about last issue? Basic usage would cost $2K+ per month. On top of the Notion subscription. On top of the AI add-on.

It's not about the money. It's about value. Claude Code costs me $200 a month and does more than all three of those agents combined.

I'm keeping the core agents: SDR, AE, CRO. They work well for what they do. But the nice-to-have agents I built "for the sake of building"? Gone. Time to be honest about what earns its keep.

Flying out 5th April, and here’s the itinerary:

San Francisco (Apr 5-11) — I'll be at HumanX conference. If you're in SF that week, lunch or coffee is on me. I want to ride a Waymo, eat at that robot burger place, and maybe sneak up to Lake Tahoe for a day. Oh, and if someone has a spare Coachella ticket for weekend one, I wouldn’t say no!

Atlanta (Apr 12-18) — Client work with PE firms and an energy company. I'm also hosting a founder and exec dinner about "AI in 2026 and what's coming." Small group, good food, and deep conversation.

If you know a good party in Atlanta, invite me. I'll DJ.

Reply to this email if you want to meet up in either city.

First come, first served on the dinner seats.

I'm about to train 1,000 people at one company. Close to 2,000 at another. We're launching behavior change programs. I'm recruiting consultants, trainers, and possibly AI engineers.

What we thought was three years away came in one.

This year is about growth and scaling what works. The context files, the clean stack, the focused agents. It's Human + AI at its best. Every week the system gets a little smarter because the context gets a little richer.

If you haven't started your context files yet, today is the day.

All you need is three paragraphs about your company and three paragraphs about yourself. You'll thank yourself next week.

Hit reply if you’ve got any questions about building your business context, and what to put in your files. I’m always happy to help.

Tim