Andrejs Karpovs's LinkedIn Strategy, Building high-agency…

Claude Cowork is amazing. If you enjoy watching your files get leaked. And If someone tells you: “You’re missing out if you’re not using Claude Cowork” They’re right. You’re missing out on bugs, ex…

LinkedIn post image: Claude Cowork is amazing.

9931933 viral

AI Security9 months ago

View on LinkedIn

One of the most powerful ChatGPT features no one talks about (once you use it, you’ll never go back): It’s called “study and learn” mode. (link in the comments) It doesn't give you all the answers.…

LinkedIn post image: One of the most powerful ChatGPT features

6905 viral

AI Education9 months ago

View on LinkedIn

December was packed with AI research. I’ve summarized 10 key papers you should read: (check attached slides for more details) 1. AI teams ≠ always better -) Agent teams help only when tasks a…

17024 viral

AI Research9 months ago

View on LinkedIn

If you’re building with agents or LLMs… You’ll want to keep this saved. LLM memory still fails. No matter how smart the model is. So here's the ultimate guide to address that. → A full map of lon…

LinkedIn post image: If you’re building with agents or LLMs…

14124 viral

LLM Engineering9 months ago

View on LinkedIn

What Zach calls an “extreme T-shaped professional” is basically what I mean when I say AI Generalist (or what I call: "GenAIralist"). I agree with the core idea: the moat in AI era increasingly comes…

9112 viral

AI Career Advice10 months ago

View on LinkedIn

There are now more AI opinions than AI models. And most of them are fine-tuned on ego. Let me break it down: We’ve got: The Doomers: “AGI is coming. We’re all cooked. Might as well unplug your GPU…

LinkedIn post image: There are now more AI opinions than AI models.

8202 viral

AI Discourse9 months ago

View on LinkedIn

Topics & Content Focus

Primary Topics

Pragmatic AI adoption for knowledge work (how to actually use LLM features day-to-day)AI security & prompt-injection risk awareness for file-connected assistantsLLM/agent engineering patterns (memory, tooling stacks, architectures)Research-to-practice translation (paper digests with actionable takeaways)

Secondary Themes

AI community discourse mapping (doomers/builders/skeptics, etc.)Learning systems and metacognition (Socratic tutoring, testing yourself)Developer productivity and automation workflows

Industry Focus

Applied GenAI for builders (LLM/agent developers, AI engineers, product builders)Enterprise AI risk, governance-adjacent security hygieneAI tooling ecosystem (open-source repos, benchmarks, infrastructure)

Content Categories

Tool tutorial / feature spotlightSecurity PSA / vulnerability breakdownBuilder playbook / reference mapResearch roundup / slide summaryOpinionated meta-commentary on the AI landscape

Performance Insights

35.16666666666666%

Avg Engagement Rate

STABLE

Performance Trend

Best Performing Topics

AI security failures in file-connected assistants (prompt injection/exfiltration)Concrete tool risk guidance with clear do/don’t recommendationsPractical, immediately-usable AI features framed as learning/productivity wins

Virality Signals

Outrage/relief trigger via security risk framing ("files get leaked")Specific mechanism walkthrough (step-by-step exploit chain increases credibility and shares)Actionable boundary-setting (use for pet projects, not enterprise) that readers can repeatRepost prompts used on resource posts to convert utility into sharesComment-link gating to concentrate discussion and clicks

Structure & Quality

190

Avg Length (Words)

HIGH

Depth Level

ADVANCED

Expertise Level

0.78/10

Uniqueness Score

Common Hooks

Contrarian/underreported benefit hook ("no one talks about")Negative sarcasm hook (praise → immediate reversal to highlight risk)Authority framing ("ultimate guide", "cleanest overview")Pattern-labeling hook (categorizing groups/types to invite self-identification)

Common Endings

P.S. question to invite repliesP.S. qualifier/caveat extending the claim (risk applies elsewhere)Soft community nudge (repost/save/bookmark for others)Link placement in comments to drive curiosity and clicks

Value Delivery Methods

Actionable playbooks (exact phrases to use, step sequences, do/don’t guidance)Curated resources (repos, paper links, slides) that reduce research timeMechanistic explanations (how the system fails, not just that it fails)Tooling landscape mapping for builders (categorization by architecture/use case)

Formatting Style

Short lines and high whitespace for scannabilityBulleted lists with arrows and hyphensPunchy one-liners followed by explanation blocksCapitalization for emphasis ("YOUR brain")Parenthetical asides and comment-link teases

Audience & Tone

YES

Question Usage

0.6%

Response Rate

Detected Tone

Smart-casual technical authoritySkeptical pragmatist (utility-first, anti-hype)Security-conscious and mildly cynicalTeacher-curator (explains + hands you the resource)semi-formalsecond-person

Interaction Style

Invites quick self-placement opinions ("where do you stand")Asks for experiential replies ("have you tried this")Encourages network-help behaviors (save/repost) rather than hard sellingUses polarizing warnings to spark corrective/additional examples in comments

Community Building Signals

Builder-first positioning (speaks to "AI builders" directly)Shared reference artifacts (repo maps, paper summaries) to become a go-to curatorLight conversational P.S. prompts to maintain a recurring comment ritual

Writing Style Patterns

Content Strategy

Hook: Contrarian/underreported benefit hook ("no one talks about")Tone: semi-formalCTA: Direct instruction CTA ("Enable X and say…")

Writing style breakdown

Most people use AI like a vending machine.

Type prompt.

Get output.

Move on.

That’s fine.

But if you’re building anything serious with LLMs…

this one habit will save you weeks.

It’s called a ‘prompt journal’.

(no, it’s not another productivity gimmick)

It’s just a simple file where you track

-) what you asked

-) what the model answered

-) what actually worked

-) what failed (and why)

-) the final version you shipped

Here’s why it matters

When an LLM fails, it fails in patterns.

Same blind spots.

Same missing assumptions.

Same “sounds right” hallucinations.

And if you don’t write those down…

you’ll debug the same thing again next week.

Andrejs Karpovs

Warm Analysis

Performance Overview

Top Posts by Engagement

Posting Patterns & Frequency

Best Performing Days

Best Performing Times To Post