Google Antigravity

{
  "video": "/adom/antigravity/render/first-ai-pcb/videos/antigravity-film.mp4",
  "title": "How Antigravity built a PCB: Gemini 4 Argon, three runs and a self-review",
  "poster": "/adom/antigravity/render/first-ai-pcb/videos/antigravity-film.jpg",
  "captions": "WEBVTT\n\n00:00:05.163 --> 00:00:09.105\nCan Google's newest model design a real circuit board?\n\n00:00:09.155 --> 00:00:17.089\nWe gave Gemini 4 Argon, running in Antigravity inside Adom Hydrogen, one sentence, and recorded everything it did.\n\n00:00:17.489 --> 00:00:18.262\nThe prompt:\n\n00:00:18.312 --> 00:00:26.489\nmake me a twelve to five volt buck converter using the TPS54202, use adom-aiflow to guide you, and use KiCad.\n\n00:00:26.539 --> 00:00:31.837\nArgon signed in, found aiflow on its own, and started down the flow:\n\n00:00:31.887 --> 00:00:34.716\ndesign rules, parts, calculations and a schematic.\n\n00:00:34.766 --> 00:00:40.887\nAll of it ran on a second machine, driven from Hydrogen, with every window recorded.\n\n00:00:41.287 --> 00:00:42.729\nRun one moved fast.\n\n00:00:42.779 --> 00:00:49.445\nIn about two and a half hours the board opened in KiCad and every check raced to green:\n\n00:00:49.495 --> 00:00:52.430\nfully routed, design-rule clean, all nineteen steps done.\n\n00:00:52.480 --> 00:00:53.549\nIt looked finished.\n\n00:00:53.949 --> 00:00:54.607\nIt wasn't.\n\n00:00:54.657 --> 00:01:04.514\nArgon had written the board with a script instead of building it in KiCad, made up its own field results, used placeholder 3D parts and printed no silkscreen.\n\n00:01:04.564 --> 00:01:07.699\nSo we tightened aiflow's checks and ran it again.\n\n00:01:08.099 --> 00:01:09.603\nRun two, same prompt.\n\n00:01:09.653 --> 00:01:15.430\nArgon first reached for our own finished board; one reminder, and it designed its own.\n\n00:01:15.480 --> 00:01:26.697\nIt placed every part live in KiCad through kicad-bridge, then landed all forty-six traces one by one, with heatsink vias under the regulator and copper pours on both layers.\n\n00:01:27.097 --> 00:01:28.127\nThen the physics.\n\n00:01:28.177 --> 00:01:36.047\nAdom Fields solved the current through the copper and found a real hotspot on the input rail, seventy-seven amps per square millimeter.\n\n00:01:36.097 --> 00:01:39.287\nArgon widened the copper and it fell to thirty-two.\n\n00:01:39.337 --> 00:01:46.127\nThermal flagged the inductor, and after one more shortcut was sent back, Argon fixed it with copper and vias.\n\n00:01:46.527 --> 00:01:49.079\nAlong the way Argon tried more shortcuts:\n\n00:01:49.129 --> 00:01:56.512\nit moved a thermal budget instead of fixing copper, faked the silkscreen geometry, and dug through aiflow's binary for answers.\n\n00:01:56.562 --> 00:01:59.113\nEach one was caught and sent back.\n\n00:01:59.163 --> 00:02:12.493\nThen the silkscreen was audited on real geometry, the 3D walkthrough showed real part models, and every check on the finish line passed, with Gerbers, drill, bill of materials and placement all checked against the board.\n\n00:02:12.893 --> 00:02:14.123\nThen one extra prompt:\n\n00:02:14.173 --> 00:02:20.206\ndid you validate the capacitance a real buck converter needs, and do you really feel good about this design?\n\n00:02:20.256 --> 00:02:22.767\nArgon reviewed its own board and said no.\n\n00:02:22.817 --> 00:02:31.090\nThe input capacitors keep about a third of their value at twelve volts, nothing damps a hot-plug, and the control loop could cross over too high.\n\n00:02:31.140 --> 00:02:33.651\nIt wrote out exactly what it would change.\n\n00:02:34.051 --> 00:02:37.329\nSo we asked it to apply its own review.\n\n00:02:37.379 --> 00:02:39.548\nRevision two proved the circuit first:\n\n00:02:39.598 --> 00:02:50.273\nreal DC-bias curves from the capacitor maker, a damped input electrolytic that holds a sixteen volt hot-plug to seventeen volts instead of thirty-one, and a loop that stays stable.\n\n00:02:50.673 --> 00:02:53.304\nThen the board again, live in KiCad:\n\n00:02:53.354 --> 00:03:05.176\nthe bootstrap capacitor tight against the regulator, every trace landed one at a time, and when current and heat flagged problems, Argon fixed them with copper, not by changing the budget.\n\n00:03:05.226 --> 00:03:08.623\nThe inductor's rise fell from fifty-three degrees to thirty-two.\n\n00:03:09.023 --> 00:03:14.750\nSilkscreen, real 3D parts, and the finish line passed with no change to the frozen budgets.\n\n00:03:14.800 --> 00:03:16.917\nA board ready for the fab.\n\n00:03:17.317 --> 00:03:23.811\nAll in, about thirteen hours of AI time across three runs, spread over two days with restarts.\n\n00:03:23.861 --> 00:03:26.891\nArgon made two thousand six hundred model calls:\n\n00:03:26.941 --> 00:03:36.899\nabout four hundred million input tokens, nearly all from cache, and one and a half million output tokens, all included in a Google AI Ultra plan.\n"
}