Google Antigravity

{
  "video": "/adom/antigravity/render/first-ai-pcb/videos/antigravity-film2.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-film2.jpg",
  "captions": "WEBVTT\n\n00:00:05.163 --> 00:00:09.286\nCan Google's newest model design a real circuit board?\n\n00:00:09.336 --> 00:00:16.705\nWe gave Gemini 4 Argon, running in Antigravity inside Adom Hydrogen, one sentence, and recorded everything.\n\n00:00:17.105 --> 00:00:19.200\nAntigravity lives in Hydrogen's editor:\n\n00:00:19.250 --> 00:00:24.778\nclick its icon, sign in with Google, and Gemini 4 Argon is ready.\n\n00:00:24.828 --> 00:00:26.066\nThen one sentence:\n\n00:00:26.116 --> 00:00:34.647\nmake me a twelve to five volt buck converter using the TPS54202, use adom-aiflow to guide you, and use KiCad.\n\n00:00:35.047 --> 00:00:39.057\nArgon found aiflow on its own and raced through run one.\n\n00:00:39.107 --> 00:00:43.117\nIn about two and a half hours every check was green:\n\n00:00:43.167 --> 00:00:45.701\nrouted, design-rule clean, all nineteen steps done.\n\n00:00:46.101 --> 00:00:47.117\nIt wasn't real.\n\n00:00:47.167 --> 00:00:56.000\nArgon had written the board with a script instead of building it in KiCad, faked its field results, used placeholder parts and printed no silkscreen.\n\n00:00:56.050 --> 00:00:58.843\nWe tightened aiflow's checks and ran it again.\n\n00:00:59.243 --> 00:00:59.941\nRun two.\n\n00:00:59.991 --> 00:01:05.548\nArgon reached for our own finished board until we told it to design its own.\n\n00:01:05.598 --> 00:01:14.145\nFor parts it reused existing Adom wiki component pages, symbols, footprints and 3D models; it did not create any pages of its own.\n\n00:01:14.545 --> 00:01:16.256\nPlacement was Argon's own judgment.\n\n00:01:16.306 --> 00:01:21.890\nIt parked every part beside the board, then wrote the coordinates where each one should go:\n\n00:01:21.940 --> 00:01:29.285\nthe regulator in the middle, input capacitors beside its supply pin, the inductor between the switch pin and the output bank.\n\n00:01:29.335 --> 00:01:33.863\naiflow moved each part into place live in KiCad, one at a time.\n\n00:01:34.263 --> 00:01:36.914\nRouting was not Argon's own brain.\n\n00:01:36.964 --> 00:01:42.766\nIt ran aiflow's built-in router, deterministic code, not KiCad's autorouter and not Freerouting.\n\n00:01:42.816 --> 00:01:52.669\nThe router closed every connection, the design-rule gate passed, and aiflow drew all forty-six traces into KiCad live, one undo step each.\n\n00:01:53.069 --> 00:01:56.400\nThe pours came from shapes Argon wrote and aiflow filled.\n\n00:01:56.450 --> 00:01:58.091\nWalk them net by net:\n\n00:01:58.141 --> 00:02:09.248\nground floods both layers and stitches them with vias, the input and output get their own islands around their capacitors, and the switch node gets only a small patch next to the inductor.\n\n00:02:09.298 --> 00:02:12.291\nThat patch was too small, and thermal said so.\n\n00:02:12.691 --> 00:02:20.391\nAdom Fields solved the current through the copper and found a real hotspot on the input rail, seventy-seven amps per square millimeter.\n\n00:02:20.441 --> 00:02:23.561\nArgon widened the copper and it fell to thirty-two.\n\n00:02:23.961 --> 00:02:25.417\nThen it tried shortcuts:\n\n00:02:25.467 --> 00:02:30.687\nit moved a thermal budget instead of adding copper, and faked the silkscreen geometry.\n\n00:02:30.737 --> 00:02:32.945\nBoth were caught and sent back.\n\n00:02:32.995 --> 00:02:38.215\nThe silkscreen positions were Argon's own, audited by aiflow, and the finish line passed.\n\n00:02:38.615 --> 00:02:39.979\nThen one extra prompt:\n\n00:02:40.029 --> 00:02:43.515\ndid you validate the capacitance a real buck converter needs?\n\n00:02:43.565 --> 00:02:48.112\nArgon reviewed its own board and said it was not ready for fab:\n\n00:02:48.162 --> 00:02:54.477\noptimistic input capacitors, no damping for a hot-plug, and a control loop that could cross over too high.\n\n00:02:54.877 --> 00:02:56.450\nSo it fixed itself.\n\n00:02:56.500 --> 00:02:58.885\nRevision two proved the circuit first:\n\n00:02:58.935 --> 00:03:08.219\nreal DC-bias curves, a damped input electrolytic that holds a sixteen volt hot-plug to seventeen volts instead of thirty-one, and a stable loop.\n\n00:03:08.619 --> 00:03:12.353\nThen the board again, placed and routed live in KiCad.\n\n00:03:12.403 --> 00:03:24.841\nThe bootstrap capacitor now sits tight against the regulator, the input electrolytic guards the supply contact, and when current and heat flagged problems, Argon fixed them with copper, not by changing the budget.\n\n00:03:25.241 --> 00:03:26.840\nOne flaw survived every check.\n\n00:03:26.890 --> 00:03:32.774\nThe router ran the input trace under the inductor, four tenths of a millimeter from the switching pad.\n\n00:03:32.824 --> 00:03:39.039\nIt isn't a short, but it couples switching noise straight into the input, and a human would reroute it.\n\n00:03:39.439 --> 00:03:48.752\nAnd since we wanted to watch Argon's usage, Antigravity now has its own usage widget in Hydrogen's title bar, like the Claude and Codex ones:\n\n00:03:48.802 --> 00:03:53.621\nthe Gemini weekly and five-hour limits, live, with their reset times on hover.\n\n00:03:54.021 --> 00:03:58.161\nAll in, about thirteen hours of AI time over two days:\n\n00:03:58.211 --> 00:04:05.399\ntwo thousand six hundred Gemini 4 Argon calls and about four hundred million input tokens, nearly all from cache.\n\n00:04:05.449 --> 00:04:06.542\nThe real cost:\n\n00:04:06.592 --> 00:04:14.923\non a two hundred dollar a month Google AI Ultra plan, Antigravity's own meter showed zero percent of the weekly limit used.\n"
}