Everyone's an AI expert until you ask how LLMs work. Stanford has the cleanest answer I've found. Their full LLM course is now FREE. 9 lectures from Transformers to Agentic systems. If you're tryi…


LinkedIn Content Strategy & Writing Style
AI Product Manager · Agentic AI · Vibe Coding | I build with Claude & teach 70K+ to do the same | ex-Techstars founder (0→$7M), ex-AI PM (Sequoia-backed)
1 person tracking this creator on ViralBrain
Basia Kubicka positions herself as a builder-teacher for practical agentic AI, combining product leadership with hands-on “vibe coding” in Claude, Hermes, and Claude SKILLs. Her content translates complex architecture into memorable mental models, then grounds it in implementation: ReAct loops, MCP, evals, knowledge harnesses, decision models, and production failures. She stands out by pairing a large-scale educational mission with unusually candid operator lessons, from cutting a workflow from 10 hours to 2 to rebuilding a brittle weekend prototype. The distinctive intersection is AI education through real product experimentation, with strong emphasis on governance, testing, cost, and human accountability rather than hype.
91.2K
29.3K
1.0K
140
3
Everyone's an AI expert until you ask how LLMs work. Stanford has the cleanest answer I've found. Their full LLM course is now FREE. 9 lectures from Transformers to Agentic systems. If you're tryi…

Spotify cut their Claude Code token bill by 90%. Then they open-sourced how they did it. The insight underneath is almost boring: most of what a coding agent does all day isn't thinking. It's I/O:…

An IBM engineer explains Agentic Harnesses in 20 min, better than any other tutorial I've seen. Tejas Kumar builds agent tooling at IBM. His whole point: you don't fix a flaky agent with a better pr…
Agent = Model + Harness I learned it the hard way building my LinkedIn Content Agent. One year and 75K+ LinkedIn followers later, here’s the Content Harness I use for every post: Hermes runs the ha…
That's exactly how you build AI agents in 2026. Nine steps, start to finish: 1 · Define the job One weekly task. Write what goes in, what comes out, what "good" means, and what it must never do. 2…
Google's 31-day AI Agents Challenge starts today. 180k+ AI builders already joined. For FREE. Built by Google Cloud AI team. One project a day you can run in 5 minutes. It’s called the Advent of…

5.0 posts/week
Posts / Week
3
Total Posts Analyzed
HIGH
Posting Frequency
0%
Avg Engagement Rate
STABLE
Performance Trend
1800
Avg Length (Words)
HIGH
Depth Level
ADVANCED
Expertise Level
88/10
Uniqueness Score
YES
Question Usage
0%
Response Rate
Writing style breakdown
<start of post>
Most teams add more documents when their agent gives a wrong answer.
Then they wonder why it still gets the answer wrong.
The agent usually wasn't missing the fact.
It had three versions of it, no way to tell which one was current, and nobody assigned to decide.
Here is the difference in plain terms.
Can I cancel this order?
The model finds a policy from last year.
It finds a newer policy in a Slack thread.
It finds a spreadsheet where somebody changed the rule again.
All three look relevant.
So the agent picks one.
That is not a retrieval problem.
That is a decision problem.
𝗧𝗵𝗲 𝗳𝗮𝗰𝘁
The customer placed the order on March 4.
𝗧𝗵𝗲 𝗿𝘂𝗹𝗲
Orders can be cancelled until they are handed to the carrier.
𝗧𝗵𝗲 𝗼𝘄𝗻𝗲𝗿
The operations lead who is allowed to change that rule.
𝗧𝗵𝗲 𝘃𝗲𝗿𝘀𝗶𝗼𝗻
The rule that was active when the answer was given.
Without all four, the agent can return a very confident answer that nobody can defend later.
A bigger knowledge base gives it more places to look.
It does not tell it which definition wins.
→ Finds rules that disagree
→ Shows the person who owns the answer
→ Saves the evidence behind the change
→ Keeps the old version readable
→ Tells the agent which version is current
The important part is not just the document.
It is the name attached to the decision.
The system decided" is not governance.
A named person, a saved reason, and a version you can roll back is governance.
That changes the agent from a search box into something your team can actually trust.
Start with one definition that causes arguments every week.
Active customer.
Late shipment.
Approved discount.
Urgent ticket.
Ask five systems what it means.
If you get five answers, don't add another document yet.
Find the person who can say which answer is real.
What definition does your agent keep getting wrong because nobody owns it?
♻️ Repost if your team is scaling retrieval and skipping approval
➕ Follow me for more on building AI systems that hold
<end of post>
Sign in to unlock the full writing analysis
Free tools to help you write, score, and benchmark in the same style.
Other creators worth studying alongside Basia Kubicka.

Paolo Perrone
No BS AI/ML Content | ML Engineer with a Plot Twist 🥷100M+ Views 📝
33 Viral Score

Dave Yang
Head of APAC - SMB and Mid Market | 4X Pre-IPO Operator (Facebook, Snap, Reddit, Grab) | Partnerships Lead Women@LinkedIn
594 Viral Score

Neil Hoyne
Chief Strategist at Google
108 Viral Score

Lucia van den Brink ✨️
experiment (CRO) to grow + stop gut-based decisions. Helping Product build data-driven orgs with A/B testing · co-Founder Women in Experimentation · CRO Agency & Freelancer Coaching · CXL Instructor
33 Viral Score

Ton Wesseling
Conversion Hotel, Experimentation island, Experimentation Culture Awards. I've been a digital optimizer by profession since 1999 and dedicated to fueling growth in the experimentation industry by organizing these events.
118 Viral Score

Ryan Levander
Conversion-Obsessed Marketer Driving Incremental Revenue Through Paid Advertising | 9+ Years Experience | AI Operations Builder
60 Viral Score
ViralBrain plans, writes, and schedules your LinkedIn content, using official LinkedIn APIs so your account stays safe.
Write like Basia Kubicka.