Why brain exercise matters now.

The adoption signal

ChatGPT's 100-million-MAU-in-two-months number from early 2023 (Reuters/UBS) compared favourably with TikTok's nine months and Instagram's two and a half years to the same threshold. By the OpenAI DevDay in October 2025, ChatGPT had reportedly crossed 800 million weekly active users — context for how mainstream the daily-AI behaviour now is.

That speed matters because it means a behavioural change happened almost overnight. A huge share of writing, summarising, brainstorming, calculating, and even coding moved from something people did to something people delegated. The question Senwitt asks is not whether that is good or bad. It is what happens to the skills underneath when delegation becomes the default.

What the early research says

A 2025 MIT Media Lab preprint by Kosmyna et al. (arXiv 2506.08872) explored what happens when people write essays with LLM help compared to writing with a search engine or unaided. The study used EEG, recall tests, and language analysis to look at neural and behavioural differences. The LLM-assisted group showed weaker neural engagement during composition, lower self-reported ownership of the essays they had produced, and worse recall of their own arguments. The work introduces the term "cognitive debt." A 2026 commentary by Stanković et al. (Stanković critique) raises methodological objections and is worth reading alongside. Both are preprints.

Adjacent evidence from 2024-2026 points the same direction. Anthropic's own 2026 study on AI assistance and coding skill formation (Anthropic Research) found a ~17% reduction in independent skill mastery for the AI-assisted group on a single new Python library. A BCG/HBR-cited study in March 2026 (Fortune coverage) found roughly 14% of AI-heavy workers reported mental fog and slower decision making after sustained AI use during a workday. The 2024 MDPI Societies study on AI tools and critical thinking (MDPI) found an inverse correlation between AI use frequency and self-reported critical-thinking engagement, mediated by cognitive offloading. None of these are individually decisive. Together they describe a real direction. See the cognitive debt page and the AI brain fry page for full source breakdowns.

Cognitive offloading, before AI

People have always offloaded thinking — to notebooks, maps, calculators, search engines, and reminders. AI assistants are an acceleration of that pattern, not a new species of it. The phrase for it is "cognitive offloading," formalised by Risko and Gilbert in their 2016 review in Trends in Cognitive Sciences (Risko & Gilbert, 2016).

The two clearest pre-AI natural experiments are worth knowing. Sparrow, Liu, and Wegner's 2011 paper in Scienceon the "Google effect" (Sparrow, 2011) showed that when participants believed information would be available online later, they remembered the information itself less reliably — and remembered where to find it better. Dahmani and Bohbot's 2020 paper in Scientific Reports (UCL spatial memory study) compared habitual GPS users with sparing GPS users and found measurable differences in spatial-memory performance years later. The takeaway is not that GPS is bad. It is that the skill underneath a tool does atrophy when the tool does the work.

The honest framing is that offloading is not bad; it is a question of which thinking acts you still want to be able to do yourself. Senwitt's answer is "the six Skills" — and the daily Set is how you keep them in practice. See the cognitive offloading explainer for the longer argument.

Why a daily habit, not a one-off course

Senwitt is shaped like Wordle or Strava — a daily ritual that you keep doing — rather than like a brain-training program that promises improvement. The reason is simple: practice that happens repeatedly tends to keep skills in reach, and the product's promise is "practise the skills, keep using the skills." That is the entire promise. It is small on purpose.

The intellectual lineage for this framing is Anders Ericsson's deliberate-practice tradition (Ericsson, Krampe & Tesch-Römer, 1993), which has held up over thirty years of expertise research: the gains from practice are specific to what is being practised, and what matters most is structured, deliberate, repeated effort over time. That is the shape Senwitt tries to hold. It is also the shape that does not over-promise.

What this is not

Senwitt is not an anti-AI product. We use AI tools every day inside the company; the team that built this site uses AI assistants the same way the rest of knowledge work does. The product is a counterweight, not a protest. The point is not to choose between AI and your own thinking — it is to keep enough of your own thinking in active use that the choice is meaningful.

Senwitt is also not a medical product, not a clinical assessment, and not a promise to prevent cognitive decline. Those are the claims that drew the FTC action against Lumosity in 2016 (FTC press release) and the 2014 Stanford-organised consensus statement from the cognitive science community (Stanford consensus). Senwitt sits outside that category by design — see brain exercise vs brain training.