If you want alpha - find it at https://www.artisangrowthstrategies.com/

Joseph W.

Mission → 1k teams to adopt weekly prod. discovery. 📈 curr: 15/1000

I use AI tools 40h/week for real work. I share what actually works.

Most AI content is either too basic ("copy this prompt!") or too technical (47-node agent workflows). I'm in the middle — practical workflows for smart people who don't want to code.

**I post about:**
• Claude Code for non-developers
• MCP integrations that extend what AI can do
• Productivity workflows I actually use

**What I'm building:**
juttu — helping SaaS companies grow revenue from existing customers.

SWE → PM → Founder. Munich.

Niko Noll

I moved my company out of Notion.

Now my docs update themselves.

Every startup I've worked at had the same problem. You write docs during onboarding, during planning, during a sprint. Then they sit there. Nobody updates them. Six months later they're wrong and everyone knows it.

Notion made this worse, not better. Beautiful docs that are just as dead.

Last week I moved everything to Obsidian. Not because the editor is better. Because the files live on disk.

Eli Mernit put it well. The architecture of an AI agent is just two things: the filesystem as state, and Claude as the orchestrator. If your docs are files, agents can work in them.

So I added agent.md files with prompts that tell Opencode/Claude Code what each doc is for and how to maintain it. The agents clean up outdated sections. Flag inconsistencies. Update references when something changes.

The docs stay alive because somthing is actually maintaining them.

We sync them over Obsidian sync.

Most documentation problems aren't writing problems. They're maintenance problems. And maintenance is exactly what agents are good at.

What's the last internal doc you actually trusted?

P.S.
Hey, I'm Niko.
I'm building Product Analyst AI and sharing how I use AI workflows and product analytics to build better products.
I also make walkthroughs and tutorials on YouTube.
Follow along if that's your jam.

I share how I use AI to build, measure, and learn faster | Founder, Product Analyst AI

Famous VC built his own CRM with AI.

He'll be back on HubSpot next month.

Michael Bloch spent 50 hours building a command center with Claude Code. Email, calendar, CRM, the whole thing.

Uses it twice a day. Saves 15 minutes.

The build was easy. Maintenance sucked. Every new feature made the system more fragile. No one maintaining it but him and a chatbot.

Sounds familiar?
Yes, I'm looking at your 34 half backed lovable projects.

Everyone's asking "will SaaS die?" Wrong question.

The real shift is supply explosion. When building costs approach zero, every vertical that had 4 competitors gets 40. Every niche that wasn't worth a venture-backed company is now viable for a lean team with AI tools.

I'm one of those new entrants. No VC funding. Small team. Building Product Analyst AI with deep domain understanding of how tech businesses use data to make more money from existing customers.

For the consumer, this is great. More specific solutions that actually solve your problem.

For the builder, the job is the same as it ever was. Solve a specific problem deeply. But the bar is higher now.

We like it that way.

Don't just ship software that theoretically helps someone get the job done. Actually get it done for them.

Code got cheap. Solving real problems didn't.

What's the one thing your current tools still don't actually solve for you?

P.S.
Hey, I'm Niko.
I'm building Product Analyst AI and sharing how I use AI workflows and product analytics to build better products.
I also make walkthroughs and tutorials on YouTube.
Follow along if that's your jam.

Last week I killed our Notion.

Here's what actually replaced it.

It starts with one file. We call it AGENT md. It's the map of our entire company. Folder structure, naming rules, key documents, and instructions for how agents should work with all of it.

The vault is one folder per area. Growth, Product, Customers, Weekly planning. Each folder has living docs that get updated continuosly and task files that get archived when done.

Three rules keep the whole thing alive:

1. Update before create. Search for existing docs first. If it exists, update it.
2. Archive don't delete. Completed work moves to _archive. Nothing disappears.
3. Link to parent. Every doc connects to its strategy doc. More connections = better context for agents.

Start of week: update This Week md with priorities. During work: create task files, link to parent docs. After completing: update the living doc, archive the task.

AI can generate endless documents. The hard part is keeping them current so the next session builds on the last one.

Full AGENT md in the comments if you want to steal it.

Are you also struggling with so much LLM output going nowhere?
This doc structure helps us, but always happy to hear suggestions on what to imrpove.

P.S.
Hey, I'm Niko.
I'm building Product Analyst AI and sharing how I use AI workflows and product analytics to build better products.
I also make walkthroughs and tutorials on YouTube.
Follow along if that's your jam.

We stopped asking AI to query data.

We taught it to think like an analyst.

First version of Product Analyst AI was simple. Connect to your data, ask questions, get answers. SQL under the hood.

It worked for "how many users signed up last week." Fell apart for anything real.

"Which users are about to churn?" It would write a query. Count logins. Return a list. Technically correct. Completly useless.

A real product analyst doesn't start with the query. They start with context. They know your product's lifecycle. They know what an AHA moment looks like for your users. They know which behaviors actually matter vs which ones just look busy.

So we built that into the agent. A context layer.

It knows your ontology. Your activation signals. Your key-value moments. The stages users move through before they upgrade or leave.

Now when you ask "who's about to churn," it doesn't just count logins. It looks at the patterns that actually predict churn for your specific product.

Same AI model underneath. Completely different answers.

If you're building AI products, this is the real unlock. Domain context beats raw intelligence. Every time.
And if you can combine the two, sure even better!

What domain knowledge would make your AI tools actually useful?

P.S.
Hey, I'm Niko.
I'm building Product Analyst AI and sharing how I use AI workflows and product analytics to build better products.
I also make walkthroughs and tutorials on YouTube.
Follow along if that's your jam.

Karpathy called it a phase shift.

I'm old enough to know what he means.

Andrej Karpathy, ex-Tesla AI head and OpenAI founding team, wrote that he went from 80% manual coding to 80% agent coding in weeks. Called it the biggest change to his coding workflow in two decades.

I'm not ex-Tesla, nor part of OpenAI. But I relate to all of it.

I build Product Analyst AI mostly in English now. 
I describe architecture, a system.
So do coding skills atrophy? Is my ego as a Software Engineer hit?

It's true:
My fingers genuinly hesitate on syntax I used to type without thinking.

But here's the part I'd add.

The engineering job didn't shrink. It shifted up.

I'm more of an architect now than a coder. I design systems. I make trade-off decisions. I think about how components fit together. That was always part of the job as an engineer. Now it's the main part.

I don't write individual functions by hand for well-proven, well-documented patterns anymore. That's the real change.

It's not slop. It's a higher level of abstraction.

And the expansion is wild. I build things now I never would have attempted a year ago. Not just faster. Things that weren't worth trying before.

Is your role shifting up or does it feel like something else?

P.S.
Hey, I'm Niko.
I'm building Product Analyst AI and sharing how I use AI workflows and product analytics to build better products.
I also make walkthroughs and tutorials on YouTube.
Follow along if that's your jam.

Half of AI agents just write code.

Every other work is still being done manually.

Anthropic published their agent deployment data. Software engineering takes roughly half of all agent usage.

Marketing, ops, data analysis, sales? All under 10%.

Greg Isenberg on this chart: "every category under 10% is a department still running on manual effort and tribal knowledge. Turn one of those workflows into a repeatable AI agent. That's vertical SaaS 2.0."

Whatever you think about AI hype. Agents will be built for all of these segments. The opportunity is figuring out how.

This is what excites us at Product Analyst AI. How do you build an agent that acutally gets the job done in analytics? Not just querying data. Helping you find signals, improve your product, make better decisions.

Every product person should be asking this right now. What does a useful agent look like for your vertical? What work does it do?

Most of these segments are wide open.

What vertical would you build an agent for first?

P.S.
Hey, I'm Niko.
I'm building Product Analyst AI and sharing how I use AI workflows and product analytics to build better products.
I also make walkthroughs and tutorials on YouTube.
Follow along if that's your jam.

Build-measure-learn is about to close itself.

OpenAI just showed us how it starts.

A small team shipped an internal product. Zero manually-written code. A million lines. ~1,500 pull requests. Three engineers. 1/10th the time.

But the code generation isn't the story.

The agents closed their own loop. Boot the app, drive the UI, reproduce a bug, record a video of the failure, fix it, open a PR, merge. While the humans slept.

Write. Test. Observe. Fix. No human in the middle.

Right now that loop is narrow. "Measure" means test results and CI passes. "Learn" means fixing what broke.

But imagine widening it.

What if agents could see usage data? Not just "does the code compile" but "are users actually clicking this button?" What if experiement results and funnel drop-offs fed directly into the agent's context?

That's build-measure-learn fully closed. We're not there yet, honestly. But the direction is clear.

This is a scary and exciting time to build. I'm choosing exciting.

What's the first piece of your product workflow you'd want an agent to close the loop on?

P.S.
Hey, I'm Niko.
I'm building Product Analyst AI and sharing how I use AI workflows and product analytics to build better products.
I also make walkthroughs and tutorials on YouTube.
Follow along if that's your jam.

taplio

The Battle

Joseph W.

Niko Noll

Followers

Posts

Engagements

Likes

Performance ScoreNiko Noll's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Joseph W.

Niko Noll

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Niko Noll's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreNiko Noll's LinkedIn Performance