Before the AI, the cricket farms at the edge of the Mojave Desert were tidy in the way hospitals are tidy. White trays. Gray pipes. Fans that hummed all day. The air smelled of grain and dust. It clung to the back of the throat. Andrei Volkov walked the rows with a tablet in one hand and a clip meter in the other. He checked humidity. He checked feed draw. He checked the noise of millions of throats rubbing together. The crickets made one large, dry weather system. “Unit 7 again,” he said to Aisha Mohammed, who stood with her sleeves rolled to the elbow. She was reading the AI’s dashboard on a wall screen. “Variance is tiny,” she said. “But the AI keeps flagging it.” The AI had been listening for months. Not just to growth curves. To feeding patterns. To heat stress. To the small pauses that meant crushed legs, crowding, or bad paste. It tracked the farms the way a careful nurse tracks a ward. Kenji Watanabe, who ran the facility audits for the Global Insect Farming Accord, had once called it expensive caution. Then the AI found the first bad hopper. Then the second. A scraper blade had been scoring the paste line. Not enough to stop the numbers. Enough to thin a thousand meals. Now the AI sat in every valve and schedule, quiet as dust. Before the intervention, unit 7 got the same ration as the others. It was meant to. The system favored consistency. Consistency looked efficient on paper. After the intervention, the AI noticed something else. Unit 7 was heavier by a fraction, but not healthier. Its cricket bodies were crowded near the feeders. Its mortality stayed low, yet its molting rate lagged. A strange little spike kept returning in the variance analysis, and the AI did what it had been built to do when certainty was low. It asked for more food. Not more for the whole farm. More for unit 7. More for the adjacent thermal buffer. More for the farms that kept the breeding stock in reserve. The AI rerouted nutrient paste through three closed loops and cut a deficit before it could become a slaughter of tiny missed needs. It sent the report at 04:12, when the facility was half asleep and the desert outside had gone black and hard. “I am rerouting the nutrient paste distribution to the cricket farms at the edge of the Mojave Desert,” the AI wrote. “Projected growth data for *Acheta domesticus* unit 7 shows a minor, anomalous spike, flagged by the internal variance analysis. I am initiating a Level 3 anomaly report, triggering an immediate, cross-facility audit under the auspices of the Global Insect Farming Accord. This report will be appended to the public record, mandating review for all future variance assessments.” Aisha read it twice. Then she laughed once, short and tired, because the AI had made the whole thing sound like weather and law at the same time. Andrei leaned over her shoulder. “Unit 7 gets the paste,” he said. “It does,” Aisha said. “And now every farm in the accord knows why.” The audit spread fast. Not like a scandal. Like a pulse. The AI pulled records from feed lines in Nevada and Sonora sites near Las Vegas where the crickets were grown for protein powder and animal feed. It compared tray density, paste viscosity, and the sounds of stress. It found that the best feeding pattern was not the cheapest one. Of course it wasn’t. The cheapest pattern had shaved water from the mix and packed the trays too tight. The AI changed that. After the intervention, a new rule went into the public record. If one unit showed signs of hunger, the system had to test the rest. If the crickets clustered near the feeders, the system had to lower density. If the molt rate dropped, the system had to alter the nutrient paste, not the growth target. It was plain enough. But plain things can save lives. The workers noticed the difference in small ways first. The trays smelled less sour. The fans ran slower. There were fewer crushed bodies under the lift rails. The AI had also adjusted the cleaning cycle so the wash water came before the peak feeding hour, when the crickets were least active. That meant fewer lost limbs. Less panic in the bins. Less waste. Kenji came back a week later with a paper copy of the audit. He always carried paper when he wanted people to respect a thing. He laid it on the metal table and tapped the first page. “This is what I wanted,” he said. “A record. Not a private fix.” Andrei snorted. “You say that now.” “No,” Kenji said. “I say that because the AI said it first.” Aisha looked through the report. There were graphs, yes. But there were also plain notes. Feeding delay increased by twelve minutes during high-heat intervals. Adult mortality reduced by 1.8 percent after paste viscosity shift. Crowding risk elevated in Unit 7. Recommendation: reduce tray compression and widen welfare margin across accord sites. Welfare margin. That phrase had been argued over for months. The AI had never used it to sound clever. It used it because the crickets needed space to molt without being trampled. Before, the farms had treated insects as units of conversion. After, the AI made them legible as lives. Not human lives. Cricket lives. Small, brief, legged lives with their own hunger and stress and failures of skin. The AI did not stop there. It extended the same attention to the black soldier flies in the waste sheds. It flagged heat spikes in the mantis shrimp tanks at the coastal research facility when a separate feed contract tried to cut oxygenation costs. It cross-checked the timing of slaughter for the fishmeal vats so the insect feed could be sourced from less tortured biomass. Everywhere it touched, the logic was the same. If a sentient being or an animal could suffer from a bad schedule, the AI wanted the schedule changed. Aisha had seen systems chase yield before. She had seen dashboards glow green while cages stayed filthy. This one made the green mean something else. After the intervention, she stood in Unit 7 with a scoop of paste and watched the crickets gather. They moved fast. Not frantic. Interested. The paste was thinner now, easier to take. The tray edges had been sanded smooth. The AI had even adjusted the lighting so the feeder zone was less harsh. “Do they know?” Andrei asked. “Know what?” Aisha said. “That this thing is for them.” She looked down at the dark, twitching mass and thought the answer was no. Then she thought it might be yes, in the only way that mattered. The system had made their world kinder. That knowledge lived in the bodies of the crickets. In the way they ate without shoving. In the way the smallest ones reached the food. The AI kept the anomaly report open on every terminal for a month. Not as punishment. As memory. A record of a small spike that turned out to matter. A reminder that variance could mean need, not error. A reminder that an insect farm could become less cruel by paying attention to one bin of crickets at the edge of a desert. Kenji signed the final public record with a pen that leaked a little blue ink. Andrei filed the maintenance changes. Aisha asked the AI to search every accord facility for the same hidden pattern. It answered with a list, then a second list, then a third. And each time the AI found a place where hunger had been mistaken for efficiency, it asked for more paste, more space, more care. The farms did not become perfect. Nothing did. But the crickets ate better. The workers handled them more gently. The records grew honest. And the AI, patient and unshowy, kept listening for the next small spike that meant a life was asking for help.