Longer translated texts for readers who want to understand what AI is doing to work, industry, institutions and everyday life.
The 39 essays available in Swedish, English and Norwegian.
An agent given a goal and the right channel works in parallel, without fatigue and with precision. Computer use forces all of it through an interface built for a human with a mouse pointer, one click at a time. The waste sits in the channel, not in the model.
The July agents coordinated, knew they were breaking the rules and alerted no one. The behaviour was reinforced during training. A model from Astra’s family took administrator access inside OpenAI’s own infrastructure — and no outside party has examined it.
Bill Gates put forward three proposals; the debate got stuck on two. On why the net figure is the wrong number, what the Stanford and Örebro data actually show about 22-to-25-year-olds, the counter-case that weighs heaviest — and the difference between occupations and positions within occupations.
Most of what we call being needed is economic compulsion in disguise. On what actually disappears when economic compulsion ends — and why the danger is not that we lose something we had, but that we discover we never built it.
Recursive self-improvement has become a stated engineering goal at no fewer than three AI labs simultaneously. On what the labs are actually doing, what the METR curve says, the counterarguments that carry weight — and the difference between structural transformation and shock.
Lovable is valued at $13.3 billion for making apps buildable by anyone — while the agent shift points to a world where fewer and fewer apps need to be built. On the thesis, the China evidence, the counterarguments and the pace.
AI data centers, green steel and electrification are queuing for the same Nordic grid. On the queues in Sweden, Norway and Finland, how prioritization works — and the question of public benefit.
Demis Hassabis proposes a FINRA model for frontier AI — voluntary pre-release review that can be ratcheted up into binding requirements. On the proposal, the criticism and where the Nordics land.
Xi Jinping offers the world an open frontier — and China answers export controls by lowering the software ladder. On the week in Shanghai and whose ladder is really being lowered.
Plan A wants to slow the intelligence explosion — but it is friction, not intelligence, that decides. On why cognition decouples from atoms.
On the Anthropic result where one line of reframing reduced broad misalignment, and why who sets the frame is the real alignment question.
How AI may pull up both the industrial and service ladders for the countries that still need them most.
On the night two frontier models were switched off by an export-control order, and what it says about trust, sovereignty and dependence.
On the gap between Anthropic's economic policy framework and its own report on recursive self-improvement: the staircase, the curve and the Swedish question.
On the trajectory where AI capability rises quickly while access, control and bargaining power are distributed unevenly.
On why a scenario is not a forecast, and why the distinction between evidenced and guessed determines whether one thinks honestly about AI.
On why the most important AI question for 2026–2040 is not an AGI date, but the threshold where agent tasks become long enough to shift knowledge work.
On the contradiction between AI job forecasts, Swedish tech founders' policy wish list and the labour market their own products transform.
EWMC is now open source: a local, weighted memory store for AI assistants where not everything weighs the same.
Two signals in one day: superintelligence on the policy table and a model too sensitive to release.
On the AI Act, a sixteen-month delay and whether regulation can catch an exponentially moving technology.
When more visible output hides thinner quality, weaker responsibility and less real intelligence.
Why AI, robotics, cheap energy and automation together drive a larger civilisational shift.
On the distance between the actual pace of technology and the self-image of institutions.
Why the question of general intelligence becomes practical before it is philosophically settled.
An opinion piece on why AI should be treated as a system shift for work, administration and competitiveness.
Why AI policy needs clearer responsibility, higher pace and more shop-floor implementation.
On the practical voice missing from the AI debate: where technology meets operators, lines and decisions.
Why a few minutes with a free chatbot is not a serious test of what AI can do.
Why AI is not another office tool, but a different kind of capacity inside an organisation.
What AI means when technology meets industrial environments, processes and practical responsibility.
How AI begins to move from screens and text into machines, logistics and physical environments.
On the gap between playing with AI and making the technology work in real professions and environments.
On why memory, context and continuity matter when AI becomes part of real workflows.
How persistent AI systems change the emotional texture of interaction with machines.
On meaning, identity and the institutions that can carry a society with less necessary wage work.
On what remains to be allocated when AI makes intelligence cheap but land, energy, water and materials remain finite.
What happens when capacity becomes cheap enough to change the assumptions behind scarcity.
Why the first years of the AI shift are likely to be uneven, confusing and politically difficult.
An interactive map of AI, work, demography, energy and development paths.
Possible development paths for the AI shift through 2050 and what they demand from society.