Hey there👋🏾 Nice to meet you too!😊

I started writing cuz I wasn’t going to do dance trends on TikTok😀

Now, I write about actionable tips to break you into product management, my journey, lessons I learned along the way, the strategies that helped me and side projects I work on.

You may have guessed it, I'm a Product Manager at Microsoft. 

Most people start their day with a cup of coffee, but my day doesn't start if it didn't begin at the gym. 

My job as a PM is to identify a problem, ask annoying questions to understand it more and ask some more questions that help bring experiment-able ideas I can test for potential solutions.

It took me 2years to break into product management. But I believe there's a quicker path and that's why I share my story.

3 things to expect

👉🏾 Actionable PM tips 
👉🏾 Inspiring side projects 
👉🏾 Sneak peek into my mind

One of my core beliefs is that "Tech people don't have two heads". You can do whatever you put your mind to do. 

I'm on a journey inspire as many people as possible to step out of the shadows and do more than just research. 

👉🏾Follow me, If you want to follow along in the journey as I discover, build, experiment, launch and share everything I learn.

Cheers

Justice U.

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.

I tested the hottest open-source AI agent.

It wanted my email, calendar, and root access.

ClawdBot runs on your hardware. Open source. Orchestrates multiple AI agents in parallel from your own machine.

Sounds great, right?

Then you look at the setup.

Email access. Calendar. WhatsApp. Root permissions on your entire machine.

And the agents... still call remote models.

So your data leaves. Your credentials are exposed. An autonomous process gets full control of your system.

That part honestly got me kind of scared.

But here's what's actually interesting.

It's one of the first real explorations of what multi-agent orchestration UX looks like. How do you manage dozens of agents running at once? How do you even stay in control?

The bottleneck isn't the AI.
It's us. Human orchestration is the constraint nobody's designing for yet.

The security needs work. But the question it's asking?

That's the right one.

What's your take... are you the bottleneck yet? :) And if so, how are you dealing?

P.S: Clawd was renamed to OpenClaw cause Anthropic complained that Clawed is too close to Claude.

P.P.S.
Hey, I'm Niko.
I share 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.

No alternative text description for this image

Anthropic just launched a legal product.

Every B2B SaaS should be paying attention.

Not because legal tech is your space. Because a model company just decided the application layer is theirs too.

Think about what that means.

The company making the AI decided it can also make the product. No wrapper needed. No middleman. They went straight to the end user.

Will they do this for analytics? For project management? For CRM? Nobody knows. But the playbook exists now.

Meanwhile, MCP is making data portable. Users can query your product from a terminal. "Death of SaaS" is trending every other week.

Some of this is overhype. Some of it is very real. And right now, honestly, I can't always tell which is which.

We're building Product Analyst AI in the middle of all this. Some mornings it feels like we're building on quicksand. Other mornings it feels like the biggest opportunity of our careers.

Both feelings are correct.

The founders who win this moment won't be the ones who predicted everything right. They'll be the ones who kept building while staying honest about what's changing.

That energy is something else right now.

Are you more excited or more terrified? Genuinely curious.

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.

I turned screen recordings into scheduled posts.

Every word is mine. The workflow is AI.

I record screen walkthroughs and talking head videos on Tella. That's the content. My words, my takes, my experience.

Used to stop there. The recording would sit on my drive and... that was it.

Now Claude Code pulls the transcript via MCP. Reads what I actually said. Suggests LinkedIn posts based on my words, not some generated template.

Then it schedules them as drafts in Typefully. Also via MCP.

Recording to transcript to draft to scheduled. I don't copy-paste anything.

Honestly didn't expect it to work this seamlessly.

The trick? AI never writes from scratch. It only works with what I already said. That's the difference between content that sounds like you and content that sounds like... everyone else.

If you're already making videos... you're doing the hard part. The rest is workflow.

What's the one part of your content workflow you wish you could automate?

P.S.
Hey, I'm Niko.
I share 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.

"Which users are about to churn?"

I asked Claude Code. It queried PostHog.

No dashboard. No funnel builder. No segment filters. I typed a question and got an answer.

Here's the setup.

One command in your terminal:
npx @posthog/wizard mcp add

It connects PostHog's MCP server to Claude Code. Log in when prompted. Done.

Next session, I started asking.

"How many users signed up last week?"
"Which features do trial users ignore?"
"Show me activation rates by cohort."

It generates the query, runs it against your data, returns the answer.

Two minutes to set up. The interface shift is real.

But that first question? "Which users are about to churn?"

PostHog MCP can't actually answer that well.

It writes a single SQL query against your event data. Counting, filtering, grouping, yes. Perfect for quick lookups.

Churn prediction isn't a query. It's pattern recognition across behaviors over time. You need a context layer on top of raw tracking data. Events and timestamps alone don't tell you who's leaving.

MCP nails the quick "I wonder if..." question. Predictions need something deeper.

Knowing the difference is the real skill.

What's the first question you'd ask your data?

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.

AI made me question everything I knew.

Then I realized that was the point.

For years I was a happy camper. Knew what SaaS looked like. Knew how B2B products worked.

Feeling confident as a product person and founder.

Then AI hits and everything is up for grabs.

How will products look? Is SaaS dead? How do you even price this?

FOMO. Fear. Anxiety. I feel all of it. Every founder and PM I talk to feels it too.

Then it clicks.

That fretting means you're conscious things are changing.

Seeing where value gets created, figuring out how software gets jobs done for users... that's our work. Was always our work.

It's just more dynamic right now.

If you're the one feeling anxious and actively learning, you're probably the one who'll help the world figure this out.

That gives me a lot of room to grow as a product builder.

I hope it does the same for you.

P.S. I'm figuring this out alongside you. You'll shape how the greatest products of the future look. Follow along if that's your jam.

Everyone's focused on AI models. I'm watching MCP.

It's the protocol that makes AI workflows real.

Every week there's a new model. Faster, smarter, bigger context window. And yeah, that matters.

But the thing that actually changed how I work? MCP. Most people have heard of it. Almost nobody actually uses it.

Model Context Protocol. It lets Claude Code talk directly to my other tools. Tella, Typefully, data sources... all connected. No API scripts. No Zapier. No glue code.

I didn't build any integrations. I just... plugged things in.

That surprised me honestly.

But here's what got me thinking as a product person.

Your product's surface area just changed. Users won't only interact through your UI anymore. They'll reach your product through Claude, through agents, through whatever MCP client they're using.

Your MCP server might matter more than your front-end soon.

The products that win next? They'll be the ones you can plug into any workflow without thinking twice.

I try to show how I actually use this every day. Real workflows, not theory. Hopefully enough to get you to try it.

Are you thinking about your product's MCP surface area yet?

P.S.
Hey, I'm Niko.
I share 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 merged two datasets nobody combines.

It revealed which behaviors drive real LTV.

Most companies keep usage data and billing data in separate worlds.

Product team looks at feature adoption. Finance looks at revenue.
Nobody connects the two.

So you optimize for engagement metrics that have zero correlation with money.

You can often infer LTV from tracking data alone. You don't always need the billing system.

But even that, most teams skip.

Start asking better questions:

Which behavior patterns lead to highest LTV?
Which acquisition channels convert to paid fastest?
What do your best customers do in their first 7 days?

These aren't hard questions. But most teams don't have the answers.

The answers change how you build product.

We've been doing this for fast-growing startups. If you want answers to questions like these, DM me.

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.

Every startup we worked with think they have a data problem. They don't. 

They have a setup problem. The data is there, it's just not structured for the questions they're actually asking.

And AI makes this harder to see, not easier:
You type a question, get a confident answer, and assume the analysis is sound. But the quality of what comes back depends entirely on what's underneath.

There are 3 levels of data questions. 
Each one needs a different setup to actually work.

Level 1: Query.
Count, filter, group. One question, one answer.
"How many trial users converted last month?" → 1,247.

Level 2: Analysis.
Compare, correlate, find patterns across behaviors.
"What do users who convert have in common that those who don't, don't?"

Level 3: Prediction.
Score users based on behavioral patterns over time.
"Which free users are most likely to upgrade this month?"

Most teams are at Level 1 and think they're at Level 2. Because the interface feels smart even when the operation underneath is simple.

The AI isn't the bottleneck. Your setup is.

We're building the layer that gets teams from Level 1 to Level 3. That's where the real insights live.

Which level are most of your data questions at?

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.

MCP analytics has a dangerous problem.

It feels smarter than it actually is.

The interface is incredible. Ask a question, get an answer. For ad-hoc questions, this is exactly the future.

But under the hood, every MCP call is a masked API request. It writes a query, runs it, returns the result. No memory. No state. No persistence.

Real analysis doesn't work like that.

You clean the data. Then segment it. Then test specific hypotheses against those segments. Each step builds on the last.

MCP restarts from scratch every time.

"Show me weekly active users." Great, here's the answer. "Now segment that by plan type." It doesn't build on the first query. It starts over. New query. New context.

You can work around this. Tell Claude Code to write Python, save CSVs, build locally. But now you need coding skill and analytics skill. The "just ask a question" simplicity is gone.

The chat interface for data is absolutely the future. MCP as the analytical engine behind it? Not there yet.

Know what it nails. Know where it breaks.

What's the most complex question you've tried asking your data?

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.

"AI > UI" was a spicy take in 2024. In 2026 it's just... what's happening.

If your product's value lives in the interface, that's worth sitting with for a minute.

Amplitude now has an MCP server. Connect it to Claude Code and just ask questions. "Which free users are closest to converting?" No dashboard. No filters. No segments. Just the answer.

Incredible for users. Less so if your product IS the dashboard.

The tension is real.

Open your data via MCP and users query it from any AI interface. Your UI becomes optional. Your edge shrinks.

Keep it closed and you're the walled garden nobody wants.

Amplitude chose to open up. Smart move. But the value now has to shift. From presentation to curation. From "here's a dashboard" to "here's what you'd never think to look for."

Dashboards won't disappear. You'll still check the same five charts every Monday.

But the ad-hoc question, the "I wonder if..." moment... these are moving to chat. And they're not coming back.

The products that thrive won't fight AI for the interface. They'll be the data layer AI can't work without.

Would you give users MCP access to your product's data, or keep the UI edge?

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

Justice U.

Niko Noll

Followers

Posts

Engagements

Likes

Performance ScoreNiko Noll's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Justice U.

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