Adam Harasimowicz

AI, data science and machine learning in e-commerce
Product manager, lead, engineer.
ex-@Intel

[Adam Harasimowicz @adharasim]

AI Agent Evaluation & Quality Measurement

https://t.co/ItunT4zagd

#AI #agents #AIagents #QualityControl #CLEAR

[Adam Harasimowicz @adharasim]

AI Agent Guardrails II: Permissions and Governance
https://t.co/nzlsum5oU0

#AI #agents #AIagents #AgenticOps #guardrails #aisecurity #security

[Anthropic @AnthropicAI]

New Anthropic research: A global workspace in language models.

Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with.

We found a strikingly similar divide inside Claude.

[Adam Harasimowicz @adharasim]

AI Agent Strategy: Build vs. Buy & Adoption Roadmap 2026 https://t.co/9gXTcEnHN4
#AI #AIagents #agents #agentops #mlops

[Adam Harasimowicz @adharasim]

AI Agent Guardrails I: Prompt Injection Defense https://t.co/h48dqerYKZ
#AI #aiagents #promptinjection #security #guardrails

[Adam Harasimowicz @adharasim]

Think LLMs should write your AGENTS.md files? Think again. Studies show human-written specs cut median runtime by ~28% and token use by ~16% compared to AI-generated noise

Stop debugging by poking the prompt → specifications are the root of reliability:
https://t.co/agiI273R4N

[Adam Harasimowicz @adharasim]

Most #AIAgent systems fail, and it's not because of model capability. The real failure is in the surrounding verification infrastructure. Learn why standard monitors show green while decisions degrade. #AIStrategy #AI

AI Agent Failure Modes & Taxonomy https://t.co/90tfkufUvD

[Adam Harasimowicz @adharasim]

🚨 AI Agents break in production because they lack engineering rigor—not because the model isn't smart enough.

Stop building demos. Build resilient systems. 🛠️

https://t.co/4Wp8Wwg2JB

#AI #LLMs #SoftwareEngineering #RAG #LangGraph #Temporal

[Adam Harasimowicz @adharasim]

An AI agent 3 hours into a session isn't crashing. It’s just quietly failing—re-asking questions and misrouting tools

If you're building agents for production, orchestration isn't your bottleneck - fundamentals are.

https://t.co/GVNscNFbO5

#AIAgents #LLMs #AI #AIInProduction

[Adam Harasimowicz @adharasim]

Why do AI agent demos work beautifully, but production success rates collapse? 📉

It’s not your framework. It’s basic multiplication.

https://t.co/HotHcpH4pU

#AIAgents #LLMs #AI #AIArchitecture

[Adam Harasimowicz @adharasim]

The real, expensive lock-in of 2026 isn't the orchestration layer—it's your model-behavioral and prompt coupling
https://t.co/xKANSmDLuh
#AIAgents #LLM #SoftwareEngineering #LangGraph #PydanticAI #ProductionAI

[Adam Harasimowicz @adharasim]

Targeted recommendations with LLMs: How to approach personalized e-commerce https://t.co/BARiGQKcuD

[Adam Harasimowicz @adharasim]

It makes me angry when I see financial scammy ads again on @YouTube after reporting them while ago.

@Google has enough capabilities to prevent it and they fail. IMHO, they don't care about it sufficiently.

#google #ads #scam #dontbeevil

[Stanford NLP Group @stanfordnlp]

Announcing Stanza v1.0.0, the new packaging of our Python #NLProc library for many human languages (now including mainland Chinese), greatly improved and including NER. Documentation https://t.co/Lm2WdAddTz Github https://t.co/UyC24Qu5P3 PyPI https://t.co/fOtAHoeFLo (or conda)

[AI at Meta @AIatMeta]

With 4.5B parallel sentences in 576 language pairs, CCMatrix is the largest data set of high-quality, web-based bitexts for training translation models. Now Facebook AI is sharing tools for other researchers to use this corpus for their work. https://t.co/uvBbfjPTk5

[Adam Harasimowicz @adharasim]

Nicely explained subplots when working with #pandas and #matplotlib - thanks @dangerscarf

https://t.co/3C3JL1Ytuw

[Adam Harasimowicz @adharasim]

State of #AI in 2019 #report: research, industry, talent, politics and China - https://t.co/rjWcWdKH5X

[Adam Harasimowicz @adharasim]

Optimizing light-weight #NeuralNetworks semi-supervised #learning. Personally, I think that in many cases semi- learning is more practical then semi-weakly approach as it does not require having an additional #model for selecting samples for teacher pre-training

[Yann LeCun @ylecun]

Billion-scale demi-weakly-supervised learning yields record results on image and video recognition.
And it's open source.

https://t.co/mrfokhRJAz https://t.co/mrfokhRJAz

[Adam Harasimowicz @adharasim]

#technicaldebts in #MachineLearning systems - https://t.co/7bHR5kSilF #ML #AI #engineering

[Adam Harasimowicz @adharasim]

Don't forget that a good #product is about #customer, #business and #technology - https://t.co/2tk5DNJyqQ