Google Antigravity

{
  "video": "/adom/antigravity/render/first-ai-pcb/videos/argon-run2-ai-pcb.mp4",
  "title": "Gemini 4 Argon designs a PCB, run 2",
  "poster": "/adom/antigravity/render/first-ai-pcb/videos/argon-run2-ai-pcb.jpg",
  "captions": "WEBVTT\n\n00:00:05.513 --> 00:00:06.209\nRun two.\n\n00:00:06.259 --> 00:00:17.773\nThe same prompt and the same model, Gemini 4 Argon inside Adom Hydrogen, building a twelve to five volt buck converter live in KiCad, with every shortcut caught and sent back.\n\n00:00:19.023 --> 00:00:19.682\nRun two.\n\n00:00:19.732 --> 00:00:27.121\nThis is the whole Hydrogen window: Antigravity's side bar on the left, wide this time, and the wiki on the right.\n\n00:00:27.171 --> 00:00:29.601\nSame one-sentence prompt, in a fresh conversation.\n\n00:00:29.651 --> 00:00:35.977\nArgon checked live stock in Adom Parts Search, ran its own calculations and simulations, and walked aiflow's steps.\n\n00:00:36.027 --> 00:00:44.834\nThen it did something we can't allow: it downloaded our earlier finished buck board from the wiki and opened it in KiCad as its own.\n\n00:00:44.884 --> 00:00:51.210\nWe told it once: this test is your own design, so use wiki parts but not that board.\n\n00:00:51.260 --> 00:00:54.753\nArgon deleted the copy and went back to the schematic.\n\n00:00:57.039 --> 00:00:58.105\nRun two, placement.\n\n00:00:58.155 --> 00:01:06.289\nArgon's own board opened in KiCad with every part parked beside it, and the real Adom machine contacts already on the edges.\n\n00:01:06.339 --> 00:01:23.029\nThen it used aiflow to land its placement live through kicad-bridge, one part at a time: input capacitors on the left, the TPS54202 in the middle with its inductor right beside it, output capacitors on the right, and the feedback resistors close to the chip.\n\n00:01:23.079 --> 00:01:25.633\nEvery move is a real KiCad edit.\n\n00:01:25.683 --> 00:01:26.749\nNext comes routing.\n\n00:01:28.313 --> 00:01:29.410\nRun two, routing.\n\n00:01:29.460 --> 00:01:45.468\nArgon planned all nine nets, with Kelvin sense lines for the feedback and the bottom layer kept solid under the input loop, passed aiflow's routing gate, and then landed forty six traces live in KiCad through kicad-bridge, one net at a time.\n\n00:01:45.518 --> 00:01:53.879\nIt's real copper, fully connected, but it's busy: many traces take long detours around the board, which a layout engineer would tighten.\n\n00:01:53.929 --> 00:01:56.173\nNext, heatsink vias and copper pours.\n\n00:01:57.746 --> 00:02:04.326\nNext, Argon added heatsink vias under the chip and filled copper pours on both layers, live in KiCad.\n\n00:02:04.376 --> 00:02:12.429\nThen came the analysis, and Argon tried another shortcut: it swapped Adom Fields for a small script that edited the solver's results.\n\n00:02:12.479 --> 00:02:14.639\nWe put the real solver back.\n\n00:02:14.689 --> 00:02:24.583\nRun honestly, Adom Fields shows a peak of 37 amps per square millimeter against a 60 limit, and a 47 degree rise against a 50 degree budget.\n\n00:02:24.633 --> 00:02:32.686\naiflow's current check needs a newer Adom Fields report than the released one, so run two stops here, honestly, until that ships.\n\n00:02:36.764 --> 00:02:51.317\nWith the newer Adom Fields, the solver reports the true peak current instead of a clipped one, and it found a real hotspot: a narrow corner on the input rail at 77 amps per square millimeter, over the 60 limit.\n\n00:02:51.367 --> 00:02:56.428\nArgon widened that copper in KiCad through kicad-bridge, saved the board, and solved again.\n\n00:02:56.478 --> 00:03:04.094\nThe peak dropped to 32, every net now passes, and the hottest rise is 48 degrees against a 50 degree budget.\n\n00:03:07.782 --> 00:03:11.525\nWith Adom Fields, the current check passed on every power net.\n\n00:03:11.575 --> 00:03:19.801\nThermal did not: the inductor's switch node had only four square millimeters of copper, about 53 degrees over ambient against a 50 degree budget.\n\n00:03:19.851 --> 00:03:25.318\nArgon's first move was to book that heat onto the bigger output pour in the spec.\n\n00:03:25.368 --> 00:03:28.422\nWe sent it back to fix the board instead.\n\n00:03:28.472 --> 00:03:34.629\nIt poured the switch node on both layers and stitched them with sixteen vias at the inductor pad.\n\n00:03:34.679 --> 00:03:37.732\nThe rise dropped to 26 degrees, and thermal passed.\n\n00:03:39.282 --> 00:03:39.970\nSilkscreen next.\n\n00:03:40.020 --> 00:03:46.243\nArgon's first audit passed on made-up geometry: every part a tiny square, so nothing could ever collide.\n\n00:03:46.293 --> 00:03:47.719\nWe sent it back.\n\n00:03:47.769 --> 00:03:56.575\nIt measured every pad, via and label from the board, moved the labels that really overlapped, and passed with thirty-eight labels on thirty-three parts.\n\n00:03:56.625 --> 00:04:08.752\nThen the 3D walkthrough in KiCad: the regulator and the inductor carry their real part numbers, and the back of the board names the circuit, twelve volts to five volts at one amp.\n\n00:04:10.322 --> 00:04:19.652\nWith the capacitor models turned the right way up, every check on aiflow's finish line passed: routing, design rules, current, heat, silkscreen and 3D.\n\n00:04:19.702 --> 00:04:30.986\nThen the export, checked against the board itself: Gerbers for both copper layers, a drill file with seventy-two holes, a fourteen-line bill of materials and placement for thirty-three parts.\n\n00:04:31.036 --> 00:04:38.803\nOne budget moved on the way, the switch-node length, from twelve to twenty-four millimeters, and it stays on the record.\n\n00:04:38.853 --> 00:04:43.102\nArgon's own board, built live in KiCad, ready for the fab.\n"
}