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

use this weekend to setup claude code.
made you a how to.
and show you some use cases.

if you haven't already this is your chance to see what this hype wave is all about.

it's neither THE END OF ALL KNOWLEDGE WORK.
Nor is it to ignore.

I spend the last 3 months mostly in Claude Code.
Like 70% of my workday.

It requires some setup, and some getting used to.
It's a new way of working.
But damn is is powerful.

Let me know what you want me to cover next!
Or question you have.

I share the AI workflows I use to get work done | Founder @ Juttu

I got the most honest question form a busy founder friend.
What the F do people actually do in Claude Code.
I had my doubts.

Even uninstalled Claude again.
Now I live in it.

Let's
- find what actual search trends are
- identify available .com domains that can rank easily
- scrape competitors to find gaps in their content (our SEO opportunity)

What would you want me to automate next week?
This week I already promised a Tella -> claude -> LinkedIn tutorial to Louise. 

Send this to someone who doubts Claude Code.
And if you are sitting on the toilet right now (i know a lot of you are), bookmark this to watch later at your desk.

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

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

What are your (non-tech) questions about Claude Code? 

I’ll do a video on them.

Joining Louise and others in doing more video in January.
 
I've been working with Claude Code for some pretty advanced non-technical workflows over the last months.

And I can really say that it revolutionized the way I work even though that sounds hyperbolic.

I'm happy to screen share how I set them up and what they do for me.

If you're not getting into these new tools and ways of working, streamlining workflows end to end, I think you will be falling behind professionally.

The only way to avoid that is by adopting a learning mindset. There are no stupid questions so just put your question below.

Stop feeling FOMO about Claude Agents.

Start simple.  

Do the work in a normal chat first. Track repeat steps in a notebook. When you catch yourself doing the same micro workflow 5 to 7 times, that is your signal to go full agent.

Generalize on the 7th.  

Tell Claude, we have done this seven times, write a concise prompt to turn it into a sub-agent. Open /agents in Claude Code. Paste the prompt. Pick Sonnet, not Opus. Cheaper, fast, good for routine tasks.

Now ... why? ... you ask

Well, these sub-agents run in parallel, keep context clean.  

Ask the main chat to launch the sub-agent for each target. Example, competitor pricing for seven sites, churn reason summaries in a fixed format, or file to summary transforms. Claude spins parallel workers, returns only the outputs, keeps the thread unpolluted.


Have you done the same task 7 times in your ChatGpt/Gemini/Claude?
Where would this save you the most time this week?

For me it was checking competitor pricing and fencing (what features are pro/premium).

My literal prompt:
```Launch competitor-research-subagent for each of the following urls. Create a summary table in competitor-landscape file```

P.S.  

Hey, I'm Niko.  

I'm building a retention tooling SaaS in public. Talking about AI, B2B subscriptions, pricing, and all that good stuff.  

Follow along if that's your jam.

Stop feeling FOMO about Claude Agents.

Start simple.  

Do the work in a normal chat first. Track repeat steps in a notebook. When you catch yourself doing the same micro workflow 5 to 7 times, t

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.

"We can't predict when churn risk goes up."

  You can. You're just not looking.

Your users are telling you. Not with words. With behavior.

The real problem isn't missing data. It's that most teams don't have a system to read the signals in real time.

Here's a simple version you can set up this week:

Step 1: Pick your 3 core post-onboarding actions.

Not signup steps. The things activated users do regularly. Could be creating a report, inviting a teammate, running an export. Define 2-3 tracking events for each.

Step 2: Set a 14-day silence alarm.

When an account stops doing those actions for 14 days, flag it. You want those users top of mind, not buried in a dashboard.

Step 3: Reach out with outcome, not features.

Don't email them "Hey, you haven't used X in a while." Lead with the outcome they signed up for.
Remind them why those actions matter for their goal, not yours.

This won't catch everything. But it catches the obvious ones most teams miss completely.

What does your churn early-warning system look like today?

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.

Conversion signals for SaaS | Co-founder @ Juttu

People in tech keep using AI to "build me another agent." Boring.
AI is way more fun when you weaponize it.

I spent 47 minutes trying to cancel a subscription this week. 
In 2026.
I won't name names.
But I turned my frustration into a game.

Introducing The Cancel Quest.

There's a leaderboard*. 
Public shame and glory.

Current high score to beat: 10 sec.

Your move.
Come play: https://lnkd.in/e_CfPKTF

*yes it means I vibe coded an app AND hooked a database. 💅🏼 <hello world> 💅🏼

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.

"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.

2026 is the year we stop talking about agents and start using them.

Most products say “agent”
And have a sparkle emoji.

The only metric that matters is jobs completed. Outcomes for the user.

My biggest learning, the harness beats the model.

What’s a harness?
It’s everything but the model.

It’s the framework enabling them to perform complex, multi-step tasks by handling critical functions like memory management, tool integration, context handling, and workflow orchestration.

Claude Code Harness is the smartest I have used.
It plans sub steps, grabs grep and other CLI tools to fetch data, and calls MCP tools when needed. It behaves more like an orchestrator, not a chat window.

For SaaS, that means pick harnesses that close loops.
Get a job done for you.

opencode is a strong open path, still very code first. I want non coding harnesses for support and analytics!

What end to end job did an agent finish for you without babysitting? If you could get a non coding harness tomorrow, which workflow gets it?

P.S.  

Hey, I'm Niko.  
I'm building a retention tooling SaaS in public. Talking about AI, B2B subscriptions, pricing, and all that good stuff.  
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