“AI Engineer” jobs are less ambiguous than the market around them
A video says almost no “AI Engineer” job matches its title. Its source says nearly the opposite: the role is consolidating while the career entry point is getting narrower.
Archive
All stories · page 32
IRZ archive, page 32.
He verifies a 3D boolean with Lean instead of reviewing 1000 lines of AI code
A developer let agents produce a 1000-plus-line 3D intersection kernel, then reduced the human trust surface to 93 lines of specification checked by Lean.
This game agent checks the screen after the code compiles
Godogen turns a prompt into a Godot, Bevy or Babylon.js project. Its most useful detail is not code generation, but a loop that captures frames and repairs visible defects.
XY does not draw 100 million points when the screen can only show a few thousand
XY is a new Python visualization library. At large scales it does not push every point into the browser: it computes only the level of detail the current screen can display.
The agent draws first; the human moves the last arrow
YuzuDraw combines a native macOS canvas, a compact DSL and an agent skill. The agent builds the structure, then the human keeps editing the same file visually.
Open amp captures are reaching mainstream pedals
Neural Amp Modeler Architecture 2 was designed for small embedded processors. The result is moving beyond plugins and into standalone pedals and amplifiers.
A Blackwing pencil is baked at 1,000°C, then soaked in hot wax
Blackwing filmed its full pencil-making process. Behind a familiar object: two extrusion stages, a 1,000°C furnace, hot wax, grooved cedar and sometimes more than ten coats of paint.
This black T-shirt reveals an adversarial pattern when heated
CVPR 2026 researchers combine thermochromic dyes, flexible heaters and an adversarial pattern. The shirt stays black at rest, then reveals the attack when heated.
Docker puts the agent in a microVM, but it can still edit your project
Docker Sandboxes isolates the kernel, daemon, network and credentials inside a microVM. But its default mode still mounts the workspace read-write. “Sandboxed” is not one trust boundary.
Meta fits a 30B local agent model into less than 20 GB
Muse Glimmer is a 30B model built for local agents. Meta says its quantized language model fits under 20 GB and the full inference stack targets a 24 or 32 GB memory envelope.