Things worth knowing, and the machines chasing what's next.
6 sourced stories and 3 KAI Documentaries -- longer, chapter-based explorations assembled only from facts already sourced elsewhere on this site.
Stories
Short, sourced discoveries.
Things You Didn't Know
The first computer bug was an actual moth
In 1947, engineers working on the Harvard Mark II computer found a real moth trapped in a relay and taped it into the logbook, writing "first actual case of bug being found." The term "bug" for a technical fault predates this by decades, but the moth is the most literal example on record.
It's a fun coincidence more than an origin story -- but it's a real one, logbook photo and all.
Space & Beyond
Voyager 1 is still transmitting from interstellar space
Launched in 1977, Voyager 1 crossed into interstellar space in 2012 and, as of recent NASA updates, is still returning data using a radio transmitter with less power than a refrigerator light bulb.
Nearly 50-year-old hardware, still phoning home from outside the solar system -- a good reminder that reliability engineering is its own kind of achievement.
AI That Feels Like Science Fiction
AlphaFold solved a 50-year-old grand challenge in biology
DeepMind's AlphaFold predicted 3D protein structures from amino acid sequences with accuracy that had eluded biologists for decades, a problem formally posed as a "grand challenge" back in 1972.
This is the AI story that actually deserves the science-fiction comparison -- not a chatbot demo, a genuine open problem closed.
Hidden Technology
Your phone's autocorrect traces back to a 1960s typo-fixing algorithm
Early spelling-correction techniques, including the Levenshtein distance algorithm (1965) for measuring how many edits separate two strings, underpin much of the autocorrect and spell-check technology still used today.
A 60-year-old piece of math is quietly fixing typos on billions of phones right now.
Strange Science
Shakey the Robot had to think for over an hour to cross a room
Built at Stanford Research Institute between 1966 and 1972, Shakey was the first mobile robot able to reason about and plan its own actions -- but its early planning computations could take a very long time relative to today's robots.
Every humanoid robot on the Future Radar today stands on this machine's shoulders.
Unexpected Inventions
The Perceptron was built decades before it could actually be useful
Frank Rosenblatt's Perceptron (1958) was an early neural network capable of simple learned classification -- but the computing power to make deep networks genuinely useful didn't arrive until decades later.
Modern deep learning isn't a new idea that suddenly appeared -- it's an old idea that finally got enough compute.
KAI Documentary
Longer, story-driven explorations.
Every chapter here links back to a real, sourced fact -- nothing new is asserted.
KAI Documentary
The Future of Humanoid Robots
A robot that once took an hour to plan a single move is now doing paid factory work. What changed -- and how far is "paid factory work" from "walks into your kitchen"?
1
Where we started
Shakey the Robot (1966-1972) needed lengthy planning cycles just to navigate a room -- see DISC-005.
2
Where we are now
Figure AI completed an 11-month BMW pilot and has paying customers; Tesla's Optimus Gen 3 entered low-volume internal production in 2026 -- see NEWS-004.
3
Where it might go
Project KAI's own Future Radar estimates MEDIUM-HIGH confidence for verified, paid commercial deployment at scale within 1-2 years -- see FUTURE-001. This is a KAI Estimate, not a forecast.
Every major AI lab is now also trying to become a chip company. Why would a software company need to design its own silicon?
1
The cost problem
Serving frontier AI models at scale is expensive enough that Anthropic, OpenAI, and Nvidia are all racing on custom silicon in 2026 -- see NEWS-001 and NEWS-002.
2
Who's doing what
Anthropic is building an in-house chip team and exploring a Samsung partnership; OpenAI built a custom inference chip with Broadcom; Nvidia's own next-gen Vera CPU already counts Anthropic, OpenAI, and SpaceX among early users.
3
What it means
Project KAI's Future Radar rates this trend HIGH confidence, already happening now, not speculative -- see FUTURE-003.
Most AI agents forget everything the moment a conversation ends. What changes when they don't?
1
The gap today
Project KAI's own Memory Agent is real but not yet automated -- its ledger exists and is actively used, but entries are written manually by the engineering process, not by an autonomous agent (see /agents).
2
Why it's hard
Reliable long-term memory means deciding what's worth remembering, verifying it stays accurate, and never silently fabricating a memory -- the same standard this entire website holds itself to.
3
The honest roadmap
Project KAI's own Future Radar rates general-purpose AI agents handling multi-step tasks with minimal supervision as MEDIUM-HIGH, 1-2 years out -- see FUTURE-008.
Comments are not connected to a backend yet -- nothing you type below is saved, sent, or visible to anyone else. This is the real, working interface design; persistence is a future, separately-decided backend project.
What should KAI investigate next?
No comments yet -- there's nowhere for them to live. This space is ready for a real comment (name, body, reply, like, report) the moment a backend is connected.
Every fact above is independently sourced -- see each card's source link. No content here is generated live by an AI agent; a future Research Agent is Planned, not built.
No comments yet -- there's nowhere for them to live. This space is ready for a real comment (name, body, reply, like, report) the moment a backend is connected.