Articles on AI.
Essays and notes on AI: where it's actually heading, what it changes about work, and what stays human. Long-form thinking, no hot takes.
LatestThe Future of Jobs in the Age of AI: Slides, Data and Sources
The reference page for my 28 October talk. Each slide, the longer version behind it, the charts and the statistics I referenced, and where every figure comes from.
Retraining for the AI Era: What the Evidence Says Works
Retraining has a measured record, and it is narrower than the brochures suggest. Training aimed at one industry with real demand pays. Generic courses have a much weaker record, and the payoff usually starts with a dip.
Digital Employees: Where AI Agents Break in Real Operations
The pitch is a worker that never sleeps. The published record is thinner: most of the proof comes from vendors' own case studies, and the reasons projects get cancelled have little to do with what the model can do.
Which Jobs Grow in Europe to 2035, and What They Pay
Both official forecasts add jobs to 2035. The fastest growth is in care, the better pay sits a qualification or two higher up, and Romania's factories tell a story of their own.
What Klarna's AI U-Turn Teaches About Deploying AI Agents
Klarna did not switch its AI assistant off. It found out what its numbers were not measuring, which is the useful part of the story for anyone about to put an agent in front of customers.
AI Adoption in European SMEs: What Eurostat's 2025 Data Says
One European company in five used AI in 2025, and one Romanian company in twenty. Read by company size, sector and stated barrier, the same numbers tell a more specific story than the headline.
How Palantir, OpenAI and Wonderful Use Forward Deployed Engineers, and What a Smaller Company Can Borrow
The companies that made the role famous run it differently, but they agree on more than they disagree. Seven patterns from their own pages, and the smaller version of each.
Why a Mid-Sized Company Needs an Engineer on Site, Not More AI Software
Mid-sized firms buy AI the way they buy software: a licence, a login, a training video. The evidence says that is where most projects stop, and the software is not the reason.
What Is a Forward Deployed Engineer, and Why Is the Term Everywhere?
Palantir invented the job to put its engineers inside its customers. Now every AI company is hiring for it. What the role is, what it does all week, and what it is not.
ChatGPT vs Copilot for Business: How to Choose, and When Neither Is Enough
An independent way to choose between ChatGPT, Microsoft Copilot, Gemini and Claude for a European SME: start from your ecosystem and data, check the settings, run a 30-day pilot with one metric, and know when you need something else.
What Is an AI Agent? A Plain Guide for Business Owners
A plain definition, how an agent differs from a chatbot and from classic automation, six examples, and the places where agents break.
What Changes Should I Make in My Business to Use AI? The Monday List
Factories gained little from the electric motor until they rebuilt around it. What to change in your business before the first AI tool.
How Should I Structure My Company for AI? Who Owns It, and Do You Need a Chief AI Officer
Who owns AI in a 50–500-person company, what AI champions do, where the engineers sit, and the governance an SME needs after the Digital Omnibus.
AI Act Article 4: What SMEs Actually Need to Do About AI Literacy
A proportionate plan for AI Act Article 4, by role, with what to document, and the corrected dates after the Digital Omnibus: high-risk obligations for HR tools now apply from 2 December 2027.
Why Is AI Not Delivering Results in My Company? 7 Reasons and the Fix for Each
Seven reasons AI isn't paying off, each with its fix, drawn from a conversation on the REWIRED podcast and the lesson of the electric motor.
Romania as an IT Delivery Location in 2026: The Price Holds, the People Don't
What official data says about Romania as an IT delivery market: an employer's cost still about half the EU average, and a talent pool that shrank from its peak for the first time.
AI and Entry-Level Jobs: Why the First Job Is Where the Change Shows Up
The first measurable effect of AI on work is not a wave of layoffs. It is a narrower door for people looking for their first job, while the forecasts for total employment still point up.
How to Choose an AI Consultant: 10 Questions to Ask Before You Sign
Ten questions that separate an engineer from a slide salesman, what a good answer sounds like for each, and the red flags not to ignore.
What an AI Consultant Costs, and What Actually Drives the Price
No invented rates: what you are really buying when you buy AI consulting, the five factors that move the price, and how to ask for quotes you can actually compare.
How to Implement AI in Your Company: A Step-by-Step Guide
A grounded engineer's playbook for implementing AI in your company, from picking the right first use case to measuring ROI and scaling what works.
AI Automation for Business: How to Spot What Is Actually Worth Automating
An honest, engineer's view on where AI automation pays off, where it backfires, and how to pick your first project.
AI Development in Romania: Options, Costs, and How to Choose a Partner
What AI development in Romania actually involves, what it costs, and how to pick someone who will ship something that works.
What a Machine Learning Solution Actually Is (and When You Need One)
A senior ML engineer demystifies the term, sets honest expectations, and helps you tell when machine learning is the right tool.
The analog spark
Digital systems are finite by construction; the brain is analog, with an infinity of values between 0 and 1. I argue that human creativity lives in that continuum, which is why today's "AI art" is really experimentation in prompting, not the machine's own spark. Yet.
My predictions for 2026 in tech
Two bets for the year ahead: holistic models that fuse audio, vision and language, and the shift from next-token to next-action data now that we've exhausted humanity's quality text.
What stays human when AI does the work
Every job is the same loop: set an objective, break it down, judge the result. AI is brilliant at the middle. The value is moving to the edges, to defining the problem and recognizing a real answer.
The future of jobs & some jobs of the future
The gig economy didn't invent new work. It changed the nature of existing work. The next shift looks familiar with a twist: a generation of AI Trainers, teaching models to do the jobs we do today.
On intelligence, abstraction & a fir tree
Our intelligence rests on a few simple mechanisms, chief among them knowing which differences to ignore. A note on abstraction, decomposition, and a tree growing from a stump.
The misleading side of the Turing Test
Turing's imitation game tries to prove the machine's intelligence by constraining the human's. A reflection on projection, lower-dimensional shadows, and losing the test ourselves.
Video generation models are not world simulators
Sora's time-consistency looks like physics, but vision is a lower-dimensional projection. A model trained on what seems to fit learns the shadow, not the world casting it.
The responsibility of dialogue
Language models learn everything they can do from the text we wrote in public. That makes how we write in the modern town square a quiet responsibility, because the mirror only reflects what we put in front of it.
On the incompleteness of ethics
Gödel proved that a consistent system always holds truths it cannot prove from the inside. Read ethics the same way, and the longing for a higher power, and then for reason, starts to look like the same unfinished move.
How a neural network learns
Open the black box of a neural network and "learning" turns out to be something humbler: following a compass of tiny nudges downhill, where reaching the true bottom sometimes means climbing a little upward first.