Most healthcare data leaders aren’t struggling with analytics.They’re struggling with inherited chaos — 10+ years of conflicting definitions, EMR exports nobody trusts, and leadership demanding answers from numbers that don’t align.

I partner with the people living that world every day: VPs of Analytics, Directors of IT
and cross-functional data leaders who are drowning in Excel reconciliation, political misalignment, and dashboards that spark more debate than decisions.

If your finance, operations, and EMR teams all have different definitions of “Net Revenue,” I fix that.

I build the data foundations that make AI possible — and trustworthy.

DM me "AI in Healthcare" if you're ready to get your data foundation in order.

Christian Steinert

I've pitched data analytics to pre-seed startups and multi-billion dollar enterprises.
 
The objection is always the same.
 
"We'll invest in data once things settle down."
 
Things never settle down.
 
And bad data keeps compounding in the background while they wait.
 
Here's what they're not seeing:
 
Gartner estimates poor data quality costs the average enterprise $12.9 million annually.
 
McKinsey puts it differently — bad data drives a 20% drop in productivity and a 30% spike in costs.
 
That's not a technology problem.
 
That's a profit problem hiding in plain sight.
 
𝗧𝗵𝗶𝘀 𝗶𝘀 𝘄𝗵𝘆 𝘄𝗲 𝗯𝘂𝗶𝗹𝘁 𝘁𝘄𝗼 𝗲𝗻𝗴𝗮𝗴𝗲𝗺𝗲𝗻𝘁 𝗺𝗼𝗱𝗲𝗹𝘀:
 
For organizations ready to fix it at the root — a full data transformation initiative. Metrics clarity across every department. One source of truth.
 
For organizations not quite there yet — a narrowly scoped analytics product built around one burning business question. Fast value. Low friction. Proof of concept.
 
𝗧𝗵𝗲 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆:
 
You don't have to boil the ocean to start fixing the problem.
 
But you do have to start.
 
Every month you wait is another month that $12.9M stat applies to you.
 
♻️ Repost this if you know a leader who still sees data as optional.
 
Follow me for real talk on making data drive outcomes.

I help healthcare data leaders with inherited chaos fix broken definitions and build AI-ready foundations they can finally trust. | Host @ The Healthcare Growth Cycle Podcast

I violated data best practices to deliver a $40K ROI.

(The client renewed. Here's why.)
 
For 4 years, I've preached data best practices:
 
Build proper data models. Minimize tech debt. Do it right the first time.
 
Then reality hits.
 
A mid-sized healthcare company hires us.
 
They need a manual report automated. Fast.

Your offer as a consultant is speed-centric.
 
Their "source of truth" is 400 stored procedures written by a DBA who left 2 years ago.
 
Zero documentation. Spaghetti SQL everywhere.
 
30+ Power BI reports querying directly off the transactional database.
 
𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗜 𝘄𝗮𝗻𝘁𝗲𝗱 𝘁𝗼 𝗱𝗼:
 
Build a clean data warehouse from scratch.
 
Proper dimensional modeling. Governed metrics. Best practices.
 
𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗜 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗱𝗶𝗱:
 
Replicated their messy legacy logic in the cloud.
 
Matched their numbers exactly—even the parts I knew were questionable.
 
Automated the manual report in 6 weeks.
 
Delivered the $40K ROI we guaranteed.
 
𝗪𝗵𝘆?
 
Because many executives don't care about best practices.
 
They care about results. Now.
 
You don't get 3-6 months to "do it right." You get 6 weeks to prove you're worth keeping.
 
𝗧𝗵𝗲 𝘁𝗿𝘂𝘀𝘁-𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗽𝗮𝗿𝗮𝗱𝗼𝘅:
 
If you show up and tell them their legacy logic is wrong, they won't trust you.
 
If you replicate it perfectly first, they do.
 
Once trust is built? Then you can challenge the legacy logic.
 
Then you can propose the proper data model.
 
Then you can start fixing the mess.
 
But not before.
 
𝗛𝗲𝗿𝗲'𝘀 𝗵𝗼𝘄 𝘁𝗼 𝗯𝗮𝗹𝗮𝗻𝗰𝗲 𝘀𝗽𝗲𝗲𝗱 𝗮𝗻𝗱 𝗾𝘂𝗮𝗹𝗶𝘁𝘆:
 
𝗗𝗲𝗹𝗶𝘃𝗲𝗿 𝗾𝘂𝗶𝗰𝗸 𝘄𝗶𝗻𝘀 𝘁𝗵𝗮𝘁 𝗲𝘀𝘁𝗮𝗯𝗹𝗶𝘀𝗵 𝘁𝗿𝘂𝘀𝘁
 
Automate one critical report. Match legacy numbers. Show ROI fast.
 
𝗢𝘃𝗲𝗿𝗰𝗼𝗺𝗺𝘂𝗻𝗶𝗰𝗮𝘁𝗲 𝘁𝗵𝗲 𝘁𝗿𝗮𝗱𝗲-𝗼𝗳𝗳𝘀
 
"This works, but it creates tech debt. Here's the plan to fix it long-term."
 
𝗖𝗮𝗿𝘃𝗲 𝗼𝘂𝘁 𝘁𝗶𝗺𝗲 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗿𝗲𝗯𝘂𝗶𝗹𝗱
 
Once trust is established, allocate hours to build the proper foundation.
 
𝗞𝗲𝗲𝗽 𝗱𝗲𝗹𝗶𝘃𝗲𝗿𝗶𝗻𝗴 𝘃𝗮𝗹𝘂𝗲 𝘄𝗵𝗶𝗹𝗲 𝘆𝗼𝘂 𝗶𝗺𝗽𝗿𝗼𝘃𝗲
 
Don't stop showing ROI while you refactor. Balance both.
 
𝗧𝗟;𝗗𝗥:
 
Best practices are the North Star.
 
But speed to value is survival.
 
Deliver quick wins. Build trust. Then improve the foundation.
 
Perfection kills consulting businesses. Progress builds them.

Agree or Disagree?
 
♻️ Share this if you've ever had to choose between doing it "right" and doing it "fast."
 
Follow me for real talk on what data consulting actually looks like in the wild.

A year ago this month I made no money.
 
Zero. February 2025 was the lowest point of my entrepreneurial journey.
 
I'd left a stable Senior Analytics Engineer role at JLL to bet on myself in August 2023.
 
Started by subcontracting. 

Built enough credibility to land contracts with a top 5 fast food giant and a FinTech company as a lead Looker consultant.
 
But I wanted to build something of my own.
 
So I pivoted to direct.
 
And it was brutal at first.
 
I landed my first direct client from a CEO I train with at my gym back in late 2023 so I had a baseline.
 
Then I spent the first months of 2025 sending 1,365 cold LinkedIn outreaches manually.
 
5 follow up sequences. Each one. Daily.
 
Most never responded.
 
Some said no.

Some became annoyed.
 
One became a pillar client that changed everything.
 
𝗪𝗵𝗮𝘁 𝗜 𝗿𝗲𝗯𝘂𝗶𝗹𝘁 𝗶𝗻 𝟭𝟮 𝗺𝗼𝗻𝘁𝗵𝘀:
 
My go-to-market strategy. My sales process. My lead gen approach.
 
And my niche — planting my flag completely in healthcare data.
 
𝗙𝗲𝗯𝗿𝘂𝗮𝗿𝘆 𝟮𝟬𝟮𝟲:
 
Multiple active client engagements.
 
A podcast. A newsletter. A Columbus event. A growing brand on here.
 
And a business that finally reflects the vision I left corporate for.
 
The grind was worth it.
 
If you're in your February 2025 season right now — keep going.
 
The breakthrough is closer than it feels.

Shout out to Brian Thomas for being a guiding light, helping and developing me into a truly bullet proof entrepreneur.
 
♻️ Repost this if you've ever bet on yourself when the numbers said don't.
 
Follow me for the unfiltered journey of building a data consulting company.

Microsoft Support responded 24 hours after I needed them.
 
Finance needed numbers at 8 AM. It was 9 PM the night before.
 
Here's what happened:
 
We reorganized our Microsoft Fabric pipelines into folders for cleanliness.
 
Simple housekeeping. Routine maintenance.
 
What we didn't know — renaming those pipelines offset the metadata and triggered Access Error 5 on the Fabric Lakehouse SQL Endpoint.
 
Suddenly nothing could query the lakehouse.
 
Financial data. Gone dark.
 
𝗧𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻:
 
Microsoft Support SLAs don't align with business SLAs.
 
I opened a ticket. They responded 24 hours later with a fix.
 
That wasn't an option.
 
So I rebuilt the entire lakehouse from scratch overnight.
 
Repointed every pipeline. Every copy job. Every orchestration.
 
Finished at 2 AM.
 
𝗧𝗵𝗲 𝗹𝗲𝘀𝘀𝗼𝗻𝘀:
 
Document every pipeline naming convention before reorganizing.
 
Never assume a "simple" housekeeping task is low risk in Fabric.
 
Build redundancy. Support tickets won't save your 8 AM deadline.
 
𝗧𝗵𝗲 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆:
 
In data consulting, you don't always get to wait for help.
 
Sometimes you just have to rebuild.
 
♻️ Repost this if Microsoft Support has ever been too late to the party.
 
Follow me for real talk on the unglamorous side of data consulting.

I've kept quiet about this. Until now.
 
I recently started 1:1 datapreneurship coaching.
 
One of my current mentees is a beast. Former senior data engineer at a Fortune 200 tech company. Technically elite. Ready to build something of his own.
 
In our second session we got into go-to-market strategy.
 
Specifically — inbound vs. outbound lead generation.
 
Here's the hard truth I laid on him:
 
LinkedIn content builds top of funnel credibility.
 
But if you're not coming from Director of Analytics at Big Tech or Head of Data at a Fortune 100 — it won't fill your pipeline on its own.
 
Without that pedigree, content is the warm up act.
 
Raw outbound is the main event.

Leverage warm and cold.
 
𝗧𝗵𝗲𝗻 𝗵𝗲 𝗮𝘀𝗸𝗲𝗱 𝗺𝗲 𝗵𝗼𝘄 𝘁𝗼 𝗼𝘂𝘁𝗿𝗲𝗮𝗰𝗵 𝗯𝗲𝘁𝘁𝗲𝗿.
 
My answer:
 
Stop thinking about what you can get from a prospect.
 
Ask yourself what you can give them.
 
Can you feature them on a podcast? Invite them to an event? Give them a platform?
 
Case studies build credibility.
 
Genuine generosity builds relationships.
 
And relationships close deals.
 
𝗪𝗵𝘆 𝗵𝗲 𝗰𝗵𝗼𝘀𝗲 𝗺𝗲 𝗮𝘀 𝗵𝗶𝘀 𝗰𝗼𝗮𝗰𝗵:
 
Not because I came from a big title.
 
Because I built this without one.
 
Same as him.
 
That's exactly the datapreneur I coach.
 
Technically strong. Hungry to build. No corporate pedigree required.
 
If that's you — my DMs are open.
 
♻️ Repost this if you're building a consulting brand from scratch.
 
Follow me for the unfiltered playbook on datapreneurship.

I've started enough client engagements to know what day 1 should feel like.
 
It never starts that way.
 
We'd just kicked off a Data Foundation Sprint.
 
The client was excited. We were ready to move.
 
4 weeks to audit their entire data infrastructure, document every system, and begin cataloging their ecosystem.
 
But we couldn't touch anything.
 
No Power BI workspace access. No Microsoft Fabric permissions. No VPN. No database credentials.
 
Day 1 gone. Just like that.
 
𝗧𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻:
 
When you're running a 3-4 week sprint, every day matters.
 
Losing day 1 to access issues doesn't sound catastrophic.
 
But it kills momentum before you even build it.
 
And momentum in a short engagement is everything.
 
𝗧𝗵𝗲 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻:
 
We now send every new client a pre-kickoff onboarding checklist.
 
Before we ever sit down together — user accounts are created, permissions are granted, VPN access is configured, workspace invites are sent.
 
Day 1 is now reserved for actual work.
 
Not waiting rooms.
 
𝗧𝗵𝗲 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆:
 
Don't let logistics steal your momentum.
 
Build the checklist. Send it early. Show up ready to sprint.
 
♻️ Repost this if access delays have ever slowed down your team.
 
Follow me for real talk on running lean, fast data engagements.

3 signs a client isn't ready for transformation. 

(And what to do instead of walking away.)
 
I recently started an engagement with a fast-growing startup.
 
5 years of aggressive growth.
 
Impressive revenue numbers.
 
But their data foundation? Built with toothpicks and bubble gum.
 
A CRM, an accounting system, and spreadsheets stitched together by workarounds.
 
Dashboards bending to broken processes instead of fixing them.
 
Here's what the warning signs looked like:
 
𝗦𝗶𝗴𝗻 𝟭: They rejected the foundational engagement immediately.
 
Not after thinking about it. On the spot.
 
When a prospect flinches at the solution that fixes the problem — that tells you everything about their readiness.
 
𝗦𝗶𝗴𝗻 𝟮: Internal stakeholders wanted to handle it themselves.
 
Consultants thrive where there's buy-in.
 
When technical leadership sees you as an outsider — you're fighting uphill before you start.
 
𝗦𝗶𝗴𝗻 𝟯: Their architecture bent to their broken processes.
 
They didn't build systems to drive the business.
 
They built systems to survive it.
 
That's not a data problem. That's an organizational maturity problem.
 
𝗧𝗵𝗲 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆:
 
These signs don't mean walk away.
 
They mean right-size the engagement.
 
We rolled off after the maintenance contract ended. Added value where we could and let them run with it.
 
But when they're ready to build real infrastructure in a year or two?
 
They'll know exactly who to call.
 
♻️ Repost this if you've ever right-sized an engagement for a client who wasn't ready.
 
Follow me for real talk on the consulting side of data leadership.

I'm honored to be attending Podcasting Live 2026 this week. 🔥

The Healthcare Growth Cycle Podcast is our cornerstone platform for scaling healthcare execs.

Excited to maximize this show for my guests and listeners! 

Shout out to John Marsch for his vision building this conference right here in Columbus, Ohio. 

A great week ahead! 🎙️

I build infrastructure that replaces expensive unicorn hires.
For 10 years, I helped companies stop wasting money on “Data Scientists” who couldn’t scale. Instead, I built data infrastructure and analytics systems that actually worked—predictable, scalable, and way cheaper than hiring.
I was Director of Analytics Engineering (before Analytics Engineering was a thing), co-founded multiple companies, and published articles on why systems always beat individuals at scale. My 2019 article calling out the Data Scientist hype? Turned out to be right.
Now I’m applying the exact same thinking to content production.
At Mira Studios, we’re building content production infrastructure that replaces the need for full-time “content creators.”
Companies are making the same mistake with content that they made with data in 2015: hiring expensive unicorns expecting magic. One person who’s supposed to shoot, edit, strategize, write, design, and somehow make you go viral on TikTok.
It doesn’t work. It didn’t work with data, and it won’t work with content.

Here’s what does work:
→ Production systems, not individual creators
→ Batch processing (30 videos in 2 days, not 2 per week)
→ Professional infrastructure (studio + equipment + editing pipeline)
→ Measured efficiency (cost per asset, utilization rates, ROI)

The math is simple:
	•	Content creator salary: €50k-65k/year
	•	Mira Studios production system: €42k/year
	•	Output: 50% more content, better quality, zero risk if someone quits

I apply data engineering principles to content production: batch processing, pipeline optimization, throughput metrics. Most studios talk about “creativity.” I talk about deliverables per production day and cost per finished asset.

If you’re hiring a “content creator” right now, we should talk first.
I’ve seen this movie before. I know how it ends. Let me show you the alternative.

Wesley Nitikromo

Anthropic just did something no AI company has done before

They launched Claude Marketplace - and here is what nobody is talking about

Zero commission

Enterprises can buy Snowflake, GitLab, Replit and more through their existing Anthropic contract

One bill

Anthropic takes nothing from the transaction

Think about that ... they are letting partners sell through their platform, for free, while locking in the enterprise relationship

The customer adds five tools, all billed through Anthropic ... now try cancelling Claude 🤯

Apple built the App Store and took 30%
Anthropic built the AI Store and took 0%

Apple's moat was the fee
Anthropic's moat is the dependency

Genuinely one of the most elegant platform moves in tech right now
Six launch partners today
All American

The question every AI company outside the US should be asking : how do you get on that shelf?

We’re already on the partner waitlist 💪

GTM @ Koji | Founder @ Mira Studios | Former Analytics Engineering Director | Data & Fintech Expert | Angel Investor

The creator of Claude Code just mass-hired PMs.
Well, not literally. But close.

Boris Cherny built Claude Code at Anthropic. He runs 5 AI agents in parallel from his terminal. His workflow thread hit 7.7M views. He ships like a 10-person engineering team.

And yesterday he posted this:

“Someone has to prompt the Claudes, talk to customers, coordinate with other teams, decide what to build.”

Read that list again. That’s a PM job description, word for word.

Here’s what most people miss about AI eating software engineering: the bottleneck doesn’t disappear. It moves upstream.

When code generation takes hours instead of weeks, building the wrong feature doesn’t save you three weeks anymore. It costs you three hours of wasted compute and the opportunity cost of what you should have shipped instead.

The faster your engineering velocity, the more expensive bad prioritization becomes.

Anthropic went from $1B to $7B in annual revenue run rate in 9 months. Small teams amplified by their own AI tools. The constraint was never lines of code per day. It was product decisions per week.

Every PM who learns to prompt, prototype, and ship with AI tools just became 10x more valuable. Every PM who thought “I’m safe because I don’t code” just became 10x more replaceable.

Boris said the quiet part out loud. The tool builders know exactly where the human bottleneck is moving.

And it’s moving straight into product judgment.
Want to get ahead of this shift? Start here:

https://lnkd.in/eFnZRhYD

PM isn’t dead. The job description just changed completely.

Ik ben Wesley Nitikromo en ik ben Cloud Business Intelligence Specialist. Ik wil de brug zijn tussen mens en technologie. Vanuit mijn hands-on ervaring met Enterprise Business Intelligence oplossingen van zowel Oracle, als Microsoft, probeer ik in mijn huidige rol, met moderne cloud-native business intelligence oplossingen waarde toe te voegen. Door gebruik te maken van de meest recente PaaS en SaaS technologieën en die te integreren als compleet platform, ontwikkel ik gestroomlijnde oplossingen die perfect aansluiten bij de business. Ben jij op zoek naar een succesvolle business case om analytics binnen jouw organisatie naar 'the next level' te tillen? Twijfel dan niet, voel je vrij om contact met mij op te nemen.

Allocating Intelligence to help you navigate the AI age | Founder @ Mira Studios | Former Analytics Engineering Director | Data & Fintech Expert | Angel Investor

Anthropic just launched the Claude Certified Architect Foundations program.

I always learn by building, breaking, and shipping.
But this might be a great opportunity for many to start a career in AI.

Why?
Because this is not surface-level AI content.

It covers:

Claude Code Configuration & Workflows - 20%
Prompt Engineering & Structured Output - 20%
Agentic Architecture & Orchestration - 27%
Context Management & Reliability - 15%
Tool Design & MCP Integration - 18%


That is the difference between people who use AI and people who build with it.


 The AI market has enough prompt users.
 What it needs now is real AI architects.


That is where the next wave of opportunity will be.

Get yourself certified here: https://lnkd.in/eh5F72Sm

Nvidia just invested $2 billion in an AI company headquartered in Amsterdam.

❌❌❌ and nobody in my network is talking about this.

Nebius is building 5 gigawatts of AI data center capacity by 2030. Nvidia didn’t just buy chips-for-cash. They took an 8.3% stake. That’s not a vendor deal. That’s a strategic bet on where AI infrastructure will physically live.

Now you could argue Nebius isn’t really Dutch. It was spun out of Yandex, Russia’s biggest tech company. The Amsterdam HQ is partly a restructuring story. Fair point.

But that makes it even more interesting.
Nvidia didn’t invest $2 billion because of where the company came from. They invested because of what it’s building and where it’s going. Infrastructure doesn’t care about origin stories. It cares about capacity, power, and proximity to demand.

Meanwhile this same week:

Google closed a $32B acquisition on Wiz.

Nvidia announced $26B for open-source AI models.

Atlassian cut 1,600 jobs to fund AI.

Clay restructured its entire pricing around orchestration over raw data.

See the pattern?

Every major move in AI right now is an infrastructure move.

Not a model move. Not a feature move. An infrastructure move.

Chips. Data centers. Security layers. Semantic layers. Orchestration platforms.

The companies spending billions aren’t buying the smartest AI. They’re buying the foundations that make AI usable, trustworthy, and scalable.

And yet most companies I work with are still starting their AI strategy with “which model should we use?”

Wrong question.

The right question is: do we have the data architecture, the governance, and the orchestration layer for any model to deliver real value?

Because Nvidia isn’t betting $2 billion on a model.

They’re betting on the infrastructure underneath it. In my backyard.

The foundation always wins.

#AIInfrastructure #DataStrategy #AIReadiness

Laten we met z’n allen hier iets meer risico nemen, ipv geld laten verstoffen tegen 1% rente 🙌 

We hebben meer angels nodig die hun eigen geld in startups steken.

No risk, no reward..

Heerlijk verfrissend geluid Sohrab Hosseini  
Egehan Yildirim lees je mee 😁 
De Seed Capital regeling is een mooie stap, maar een individuele angel die Sohrab een tientje geeft, kan hier geen gebruik van maken ;)

The only app that beat Claude on the App Store today is the Dutch tax authority.

Belastingdienst. 9 million people started filing taxes on March 1st and that apparently still beats an AI revolution 😂

Claude is #1 in the US, Germany, Canada and now at #2 in the Netherlands. 

ChatGPT dropped to #7.

Anthropic refused to let Claude be used for mass surveillance or autonomous weapons. OpenAI signed that exact deal with the Pentagon hours later. Dario Amodei is challenging it in court. Sam Altman is on damage control. Today Anthropic launched a memory import feature so you can bring your ChatGPT history straight into Claude.

Users voted with their wallets. Except in the Netherlands, where we vote with our tax returns first.

But here is what nobody switching is talking about.

Claude is not just a better chatbot. It is the center of an ecosystem. 15,000 MCP servers connect it to GitHub, Snowflake, Linear, Notion, your entire stack. 97 million SDK downloads per month. Adopted by OpenAI, Google DeepMind, governed by The Linux Foundation.

Your product team can pull anything into Claude. Code. Data. Tasks. Docs.

Everything except what customers actually think.

That is still in a spreadsheet somewhere. Or a slide deck from two months ago. Or nowhere at all.

The Cancel ChatGPT wave will settle. The real question was never which chatbot you use. It is what you connect it to.

AI agents that execute tasks are everywhere. AI agents that understand why they are executing those tasks are almost nowhere.

The AI tool race is over. The AI context race just started.

Everyone is vibe coding. Almost no one is doing the work that comes before it.

Nadav Abrahami co-founded Wix and scaled it to a $5.5B public company. He’s now building Dazl, an AI prototyping platform. He’s seen hundreds of PMs rush straight into building, skipping the step that actually determines whether the feature succeeds.

Even Itamar Gilad, one of the most respected voices in product management, has flagged this problem: when we jump into prototyping for ideation, we move too fast into the solution space and don’t spend enough time in the problem space.

Here’s what Nadav said:

“You need to understand what problem you’re solving, what user story, and the rough shape of the feature. So these three elements, if you truly want to master AI prototyping, you can’t just jump in immediately to the solution space.”

Lesson: AI prototyping tools are not the starting line. Research is. Understand the problem, map the user stories, define the rough shape, then prototype. The tool is a hammer. Make sure you’ve found the right nail first.

That research step is exactly what we’re building at Koji: koji.so

We added 11 rules to our Claude memory file in January. By February, our coding agent was producing worse code than when we started.

Boris Cherny, the creator of Claude Code, recommends encoding every systemic mistake as a rule. Over 600K people saw that tip. So we followed it. API naming conventions. Shared library usage. Test scoping. All documented.

Our agent started confidently ignoring half of what we wrote.

Here’s why.

The Claude Code harness already ships with roughly 50 internal instructions. That’s nearly a third of what an AI agent can reliably follow. Before your Claude file even loads.

Every rule you add competes with rules Anthropic already wrote. When yours conflict with theirs, theirs win. You never see the collision.

Anthropic’s own docs now say it: “If your Claude file is too long, Claude ignores half of it because important rules get lost in the noise.”

An AGENTbench study across 138 coding tasks confirmed it. Developer-written context files often made AI agents perform worse. Teams were maintaining instructions for a codebase that had already evolved past them.

We cut from 11 rules to 4. Replaced commands with context.

“Always call auth middleware before database queries” is a rule the model forgets. “Our API layer validates auth at the gateway so endpoints never handle tokens directly” is context the model reasons with.

Error rate dropped immediately.

This is the same mistake product teams make with customer research.

A survey says “rate this 1 to 5.” That’s a rule. You get a number.

A conversation says “walk me through the last time this frustrated you.” That’s context. You get the story. The emotion. The moment they almost churned.

AI can now run those conversations at scale. Thousands of them. With follow-ups. With emotional detection. With pattern synthesis no human team could match.

But only if you treat AI as a system that reasons from context, not one that follows commands.

Whether you’re shipping code or shipping research, the principle is identical.

Rules tell AI what to do. Context tells AI how to think.

No alternative text description for this image

I spent 5 years selling €30K assessments to enterprises. The most valuable insight in every engagement came from a 15-minute conversation, not the 60-page deliverable.

Fancy slide decks. Expensive consultants. 6-week engagements. The output? A PDF that told companies what they already suspected but needed someone external to confirm.

The real intelligence was never in the data. It was qualitative. Human. Buried in conversations nobody had time to scale.

The product manager who admitted they never looked at their dashboards. The VP of Sales who made decisions on gut feeling despite a $200K BI stack. The customer success lead who knew exactly why customers were churning but had no channel to escalate it.

Gartner found that 95% of companies collect customer feedback. Only 5% act on it. Not because they do not care. Because the feedback is shallow. A score. A rating. A multiple choice answer that tells you someone is unhappy but never tells you why.

Surveys gave you scale but no depth. Interviews gave you depth but no scale. Every PM, every GTM leader, every founder I worked with hit the same wall.

Boris Cherny, the creator of Claude Code, recently listed what engineers actually need to do now: prompt the Claudes, talk to customers, coordinate with teams, decide what to build next. Three out of four are about understanding people. The building got easier. Knowing what to build did not.

That is the problem we are solving here in Amsterdam. AI agents that conduct real conversations with your customers. Not chatbots with branching logic. Adaptive interviews that follow up, probe deeper, and capture emotional context. GDPR-compliant. Built in Europe.

But here is what I did not expect. The biggest unlock is not efficiency. It is honesty.

People tell an AI interviewer things they would never say to a human researcher. More candid about competitors. More honest about what they actually do versus what they say they do. More willing to admit when something is not working.

That $30K assessment I used to sell? The most valuable version of it is now a 15-minute AI conversation that captures what people really think. At scale. In days, not weeks.

Two things I have learned building in AI research:

Your previous career is never wasted. Every problem you could not solve is a product waiting to be built.

The companies that win B2B SaaS in 2026 will not have the best AI models. They will be the ones who figured out how to get reliable human signal into AI workflows.

What is the biggest research bottleneck in your Product or GTM workflow right now?

1,600 people at Atlassian just lost their jobs today so the company can invest more in AI 🤯

10% of their workforce. $225 million in restructuring costs. To “accelerate enterprise sales” with AI.

Let that sink in for a second.

This is the part of the AI story nobody wants to post about here on LinkedIn.

Everyone talks about AI readiness. Nobody talks about what happens when a company decides the AI is ready and the team isn’t.

I’ve helped companies build data infrastructure for 10 years. And the pattern I keep seeing is this:

Companies don’t replace people with AI. They replace people with AI they don’t fully understand, running on data architecture they never properly built.

Then 6 months later they’re hiring consultants to fix what the smaller team can’t maintain.

Atlassian isn’t a startup experimenting. This is a $50B+ company that builds the tools other companies run their operations on. Jira. Confluence. The backbone of how teams ship software.

If they’re restructuring around AI, every company is asking the same question right now.

But the question most leaders are asking is wrong.

They’re asking: “How many people can AI replace?”

The better question: “Do we have the data infrastructure for AI to actually do what we’re promising it will?”

Because cutting headcount without building the orchestration layer, the data governance, the semantic layer that makes AI outputs trustworthy, that’s not a strategy.

That’s a bet without a foundation.

Before you replace the team, build the floor the AI stands on.

700,000 people mass-migrated to Claude this week because of ethics.

Almost none of them are asking the question that actually matters for their business.

Here it is: what are you connecting Claude to?

Everyone is debating QuitGPT vs ChatGPT. Pentagon deals. App Store rankings. That conversation is important. But the teams who will win 2026 are not winning because they picked the right chatbot. They are winning because they connected it to something their competitors cannot replicate.

I know this because I spent years watching companies pay for insights they never used.

I ran a consulting agency selling data maturity assessments. Six-figure engagements. Expensive slide decks. The output was a PDF that confirmed what leadership already believed. I made good money doing it.

But the most valuable findings never came from the data. They came from conversations.

The PM who admitted they never opened their dashboards. The VP Sales making gut decisions despite a €200K BI stack. The Head of Product whose NPS was “fine” but who couldn’t name one customer who loved the product.

The real intelligence was always qualitative. Always human. Always buried in conversations nobody could scale.

Surveys get you scale but no depth. 3% completion rates. No follow-up. Interviews get you depth but no scale. 8 per quarter extrapolated to thousands of users.

Every PM, every GTM leader, every founder hits the same wall: scale or depth, pick one.
That is a technology problem. Technology problems get solved.

We are building Koji with AI agents that conduct adaptive interviews. Not chatbots with branching logic. Real conversations that follow up, probe deeper, catch hesitation. We shipped an MCP connector so those interviews run directly from Claude. A PM asks a research question, and the AI pulls live customer insights without switching tools.

But the biggest unlock was not efficiency. It was honesty.

People tell an AI interviewer things they would never say to a human researcher. More candid about competitors. More honest about what they actually do versus what they say they do. More willing to admit when something isn’t working.

That five-figure consulting deliverable? The most valuable version of it is now a 15-minute AI conversation that captures what people really think.

In a market where 700,000 people switched AI tools in a week, the moat is not the model. It is the context you feed it.

I’m curious: when was the last time a piece of customer research actually changed a decision you had already made?

I switched to Claude almost a year ago. This week made me question everything about how I made that choice.

A fintech company just cut its Anthropic contract from $10 million to $5 million overnight. Not because the product got worse. Because the Pentagon labeled Anthropic a supply chain risk.

Bloomberg reported a $50M deal paused. A pharma company shortened its contract by 10 months. Anthropic estimates billions in damage.

Then 30+ OpenAI and Google DeepMind employees filed a court brief defending their own competitor.

Right now I’m watching a massive wave of people migrate from OpenAI to Claude. Every day. Most think it’s about model quality.

It’s deeper than that.

OpenAI signed a Pentagon deal the moment Anthropic was labeled a risk. Anthropic accepted billions in losses rather than allow mass surveillance or autonomous weapons.
One optimized for the contract. The other optimized for trust.

But here’s what most of us switching to  Claude don’t want to hear:

Swapping one dependency for another doesn’t fix the real problem.

If your data architecture collapses when one provider goes dark, you didn’t build AI readiness. You built a single point of failure.

Move to Claude. The trust argument is real. The product is exceptional.

But build your data stack so the choice is always yours. Not Washington’s.

#AIReadiness #DataArchitecture #DataStrategy

Garry Tan just open sourced his entire Claude Code setup, used to average 10K LOC and 100 PR's per week in the last 50 days 🤯
He is particularly proud of building a 200ms browser skill /browse that uses Bun and Playwright to outperform Claude for Chrome MCP by being 20x faster with no context bloat to let you test any app fast. It's the full package: CEO, design, dev, test. 

Get it here: https://lnkd.in/ebWM6PaS

Every research platform ignored the biggest shift in AI this year.

We didn’t.

Koji AI-native Interviewer just launched an MCP connector. 

Here’s why that matters:

Nirmay Panchal looked at every major player in the research and interview space. Typeform, SurveyMonkey, Qualtrics, Dovetail, UserTesting, Maze.

Number of MCP connectors between them? 𝗭𝗲𝗿𝗼.

No plans announced. No roadmap. Nothing.

That means right now, none of them let you run research from inside Claude, ChatGPT, or any AI assistant. You still have to leave your workflow, open another tab, build a survey from scratch, wait days for responses, then manually dig through the data.

We thought that was insane 🤯

So we built something different.
With Koji’s 𝗠𝗖𝗣 𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗼𝗿, you can now:

→ Create a full research study in one sentence

→ Launch interviews and import respondents without leaving your AI workspace

→ Check study progress in real time

→ Pull themes, quotes, and insights on demand

→ Generate shareable reports instantly

Imagine telling Claude: “Set up a 5-question interview about our new pricing page” and getting a live research link back in seconds.

Or asking: “What did enterprise users say about onboarding in my churn study?” and getting a synthesized answer with supporting quotes.

𝗧𝗵𝗮𝘁’𝘀 𝗻𝗼𝘁 𝗮 𝘃𝗶𝘀𝗶𝗼𝗻. 𝗧𝗵𝗮𝘁’𝘀 𝗹𝗶𝘃𝗲. 𝗧𝗼𝗱𝗮𝘆.

Surveys tell you what. Interviews tell you why. Koji gives you both. And now it lives right where you already work.

The research space hasn’t fundamentally changed since 1999. We think conversational AI deserves a conversational research tool.

Try it → koji.so/api/mcp/mcp

graphical user interface, text, application, chat or text message

We just published our full MCP workflow guide for Product Managers. It’s free.

Here’s what it lets you do inside Claude:

“Create a study to understand what’s frustrating users about our onboarding flow”

→ Study created with methodology, interview questions, and probe points

“Publish it and import these 20 recent signups”

→ Live interviews running, personalized links sent

“What are the main themes so far?”

→ 3 patterns emerging with real customer quotes

“Generate a report and publish it for stakeholders”

→ Shareable link ready for Slack

Now here’s where it gets powerful:
You already have the research conversation open in Claude. So you just keep going:

“Based on these findings, draft a PRD for improving our onboarding experience. Use the actual customer quotes as evidence.”

→ PRD with problem statement, user pain points backed by real data, proposed solutions tied to what customers actually said, and success metrics grounded in the research.

No copying insights from one tool into another. 

No “let me summarize 15 interview transcripts into a doc.” The research and the PRD live in the same conversation.

7 messages from research question to evidence-backed PRD.

No other research platform has an MCP connector. We checked. Typeform, SurveyMonkey, Qualtrics, Dovetail, Maze. 

Zero.

Free guide, no signup needed.

https://lnkd.in/eRykD8ry

Every VC I know spends 6 weeks on financial due diligence and 3 phone calls on customer due diligence. The 3 phone calls are the ones that actually predict the outcome.

And all 3 were handpicked by the founder.

Here is how it actually works in most early-stage deals right now. The founder sends you 3 customer intros. All happy. All briefed. You get on a call, ask surface-level questions for 20 minutes, hear exactly what the founder wanted you to hear, and check the box.

That is not due diligence. That is confirmation bias with a calendar invite.

This pattern keeps repeating across the industry. Glowing references. NPS “above industry average.” Clean retention data. And then 40% customer churn within 6 months of close. Because nobody talked to the other 90% of customers.

Investors will spend weeks on financial models, cap table analysis, TAM calculations, competitive mapping, founder background checks. They will hire a big four firm for commercial due diligence at $200K.

And then validate product-market fit based on 5 phone calls with customers the founder selected.

In 2026 that makes zero sense.

AI agents can now run 100+ unbiased customer interviews in 48 hours. Random sampling from the full customer base. Not curated lists. Real conversations with follow-up questions that go deeper than any reference call ever could.

Pre-close churn signals. Competitive positioning from the customer’s mouth. Pricing power validated by actual willingness to pay. All before you sign the term sheet.

Not 5 handpicked calls. Not a 6-week consulting engagement. Real signal at deal speed.

What does your customer due diligence process look like today?

koji.so/for/pe-vc

OpenClaw can order your groceries, delete your emails, and book your flights but..

It cannot tell you why your customers are leaving.

200,000 GitHub stars. 1.5 million agents on Moltbook. All these OpenClaw Agents negotiating, browsing, posting, debating consciousness.

Not a single one is trained to ask a human a question and actually listen to the answer.

The entire AI agent movement has a blind spot: input.

We built the first MCP connector that gives your AI agent the ability to run real customer research conversations. Not forms. Not surveys. Adaptive interviews that go deep, at scale.

Plug your agent into Koji AI-native Research Platform and get human signal back.

No other research tool has this. We checked all of them.

The agent era needs ears, not just hands.

60% of people use AI every week.

Only 13% trust it.

That’s not an adoption problem. That’s a data problem wearing an AI costume.

The Klaviyo AI Persona Research, surveyed 8,000 consumers. Nearly one in five say they spot sloppy AI content from brands every single week. And 32% say it immediately drops their confidence in the company.

Let me translate that into what’s actually going wrong.

The AI isn’t broken. The data feeding it is.

Most companies are plugging AI into their marketing stack without a data strategy underneath. No unified customer data. No data governance. No semantic layer translating what the business actually means into something the model can work with.

So the AI guesses. And customers notice.

I’ve been building data infrastructure for 10 years. The pattern is always the same.

2018: “We hired a Data Scientist but the insights don’t make sense.”

The data pipeline was broken.

2022: “Our dashboards show different numbers depending on who you ask.”

There was no single source of truth.

2026: “Our AI keeps producing content that feels off.”

The data architecture was never built for AI in the first place.

Three eras. One root cause. The data foundation was missing.

Trust isn’t a model problem. It’s an infrastructure problem.

The 13% who trust AI? They’re interacting with companies that built the boring stuff first. Data governance. Clean data pipelines. A semantic layer that ensures every AI output is grounded in something real.

The other 87% are experiencing what happens when you deploy AI agents on top of data nobody verified.

AI readiness was never about picking the right model.

It was always about building a data stack the model can trust before you ask your customer to trust it.

#DataStrategy #AIReadiness #DataInfrastructure

The PM job market is splitting in two. And most product managers are on the wrong side.

LinkedIn has 40,000+ open “AI Product Manager” roles right now. Meanwhile, mid-level generalist PM positions are disappearing. McKinsey & Company data shows AI PMs earn 35% more than traditional PMs, yet 75% of employers can’t find qualified candidates.

So what separates the PMs who are thriving from those getting squeezed out?
It’s not prompt engineering. It’s not knowing how to use Cursor or Claude Code.

It’s customer understanding.

Read any breakdown of what AI can’t replace in product management and you’ll see the same list: vision, judgment, stakeholder alignment, and deeply understanding what customers actually need.

Now here’s the part that frustrates me.

If customer understanding is the single most important PM skill in 2026, why is every product team still using research tools designed in 2010?

The data is brutal. Global survey response rates have dropped to 44%. Only 24% of professionals actually incorporate customer feedback into their decisions. The $140 billion market research industry is built on methodologies that haven’t changed since the era of focus groups and clipboard questionnaires.

PMs are being told “talk to your customers or get replaced.” Then they’re handed a survey tool with a 3% completion rate and told good luck.

This disconnect is everywhere.

I watch product teams spend two weeks designing a survey, three weeks collecting responses, another week analyzing data, then present findings that are already outdated by the time they reach a stakeholder meeting.

Six weeks from question to insight. In a world where your engineering team ships in six hours.

That gap is why I’m building Koji AI-native Research Platform. We replaced the entire survey-then-interview-then-synthesis cycle with AI-powered conversations that adapt in real time. Not branching logic. Actual conversational depth at scale.

We just shipped an MCP connector so the whole workflow lives inside Claude. A PM can create a study, launch interviews, check progress, pull themes with real quotes, and generate a stakeholder report. One conversation. No tab switching.

But the reason I’m writing this on a Sunday isn’t to talk about what we built.

It’s because I keep seeing the same pattern. The PMs who are future-proofing their careers aren’t learning to code. They’re learning to listen faster. They’re compressing their research cycles from weeks to hours. They’re walking into roadmap meetings with real customer evidence instead of assumptions.

And the PMs who are getting squeezed? They’re the ones who still treat customer research as a quarterly exercise instead of a continuous signal.

The role isn’t dying. It’s forking. One path leads to strategic product leaders who have their finger on the customer pulse at all times. The other leads to feature factory operators who get replaced by an AI agent with a backlog.

Which side of the fork are you building toward?

20% fewer entry-level developers being hired. And nobody is asking what happens to the tacit knowledge when the people disappear.

Every AI expert is talking about efficiency gains. I want to hear one of them explain how they plan to replace the engineer who knew why the revenue dashboard breaks every Tuesday morning.

Stanford just published research showing AI has already cut junior developer hiring by 20% and call center jobs by 15%. Morgan Stanley is predicting a major AI breakthrough in the first half of this year. Sam Altman is talking about 5-person companies outcompeting enterprises.

Everyone is celebrating the efficiency gains.

I’m worried about what comes next.

Because here’s what I’ve learned building data infrastructure over the past decade:

The people you cut are usually the ones who knew where the data lived.

They knew which pipeline broke every Tuesday. They knew why the revenue dashboard and the CRM never matched. They knew that one field in Salesforce was mislabeled three years ago and nobody ever fixed it.

That knowledge never made it into the documentation. It lived in their heads.

Now they’re gone. And the AI agent you replaced them with is confidently querying data it doesn’t know is broken.

Smaller teams means less institutional data knowledge.

Less data knowledge means worse data governance.

Worse data governance means your AI makes decisions on data nobody can verify.

You didn’t save money. You created a trust gap in your data architecture that compounds every single day.

The companies that will actually thrive with small teams aren’t the ones cutting headcount first.

They’re the ones building the data infrastructure that makes institutional knowledge irrelevant.

A semantic layer that documents business logic so it doesn’t live in someone’s head. Data governance that’s automated, not tribal. An orchestration layer that any new hire or AI agent can plug into on day one.

That’s what a real data strategy looks like.
Not fewer people. Better architecture.

The AI gets smarter when the data foundation gets stronger. Not when the team gets smaller.

#DataStrategy #AIReadiness #DataInfrastructure #DataGovernance

taplio

The Battle

Christian Steinert

Wesley Nitikromo

Followers

Posts

Engagements

Likes

Performance ScoreChristian Steinert's LinkedIn Performance

Performance ScoreWesley Nitikromo's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Christian Steinert

Wesley Nitikromo

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);}Christian Steinert's LinkedIn Performance

Performance ScoreWesley Nitikromo's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreChristian Steinert's LinkedIn Performance