- News
- TOI Blogs News
- Digital Life News
- Neural Coding: AI system to enable software app development in your sleep
Neural Coding: AI system to enable software app development in your sleep
AI system to enable software app development in your sleep
Most of us have gone to bed with a problem and woken up with the answer. Now imagine waking to find the answer already built and waiting on your phone as a working app. That is the promise behind neural coding, a term I first used in this sense in Forbes in September 2025 to describe AI systems that turn brain signals directly into software. Part of that promise is already real. The rest is still research, and it deserves public attention before it reaches the market.
What is already real
In February 2025 the AI researcher Andrej Karpathy coined the phrase “vibe coding” for a new way of building software: describe what you want in plain language and an AI model writes the code. Collins Dictionary named it Word of the Year for 2025. Since then the tools have grown from assistants into agents. Give one a goal in the evening and it can plan, write, test and correct code for hours without supervision, then present the result in the morning. In a literal sense, software is already being developed while its owners sleep.Neural coding asks what happens when the typed instruction disappears too. The input becomes the brain itself, read by a brain-computer interface (BCI) and interpreted by AI.
Reading the brain, awake and asleep
In August 2025, Stanford University researchers reported in Cell that implanted sensors could decode inner speech, the sentences a person silently says in their head, with up to 74% accuracy from a vocabulary of up to 125,000 words. In March 2026, China’s medical regulator approved an implanted BCI for commercial use, widely reported as a world first, to help some patients with spinal cord injuries grasp objects again. In June 2026, a team led by the University of California, Davis reported in Nature Medicine that a man with the motor neurone disease ALS had used his brain implant at home for more than 3,800 hours over nearly two years, communicating at an average of 56 words per minute and holding a full-time job.Sleep science has moved too. In 2021, a Northwestern-led study in Current Biology showed that people in lucid dreams could solve simple arithmetic problems and signal the answers with eye movements. In February 2026, Northwestern researchers reported that sounds played during REM (Rapid Eye Movement) sleep nudged dreams towards unsolved puzzles. Puzzles that appeared in dreams were later solved 42% of the time, compared with 17% for those that did not.Put these strands together and the title of this article stops sounding like science fiction. I must be honest about the gaps. To my knowledge, no one has decoded a software design from a sleeping brain. The strongest BCI results come from surgically implanted devices in a small number of patients with paralysis, calibrated for each individual. Wearable headsets are far noisier. The first practical forms of neural coding will most likely serve people who are awake.
What we must be mindful about
That is why the hard questions belong to today.First, a thought is not a specification. In Communications of the ACM (August 2026) I argued that translating thought straight into software risks severing intent from implementation. Engineering lives in the steps between wanting something and shipping it: stating requirements, weighing trade-offs, testing and review. A neural interface can help collect and refine intent, but engineering governance must still decide what gets merged. This matters even more at night. A dreaming mind is associative and unguarded, and none of us would want our dreams released to the public as software.Second, frictionless tools change us. In IEEE Computer (July 2026) I described neural coding as a cognitive trap. The effort of putting an idea into words or code is where much of our understanding forms. Remove that effort and we weaken the thinking, accountability and safety that good software depends on.Third, verification. When the security firm Veracode tested code from more than 100 AI models in 2025, 45% of the results contained a known security flaw, and its 2026 update found little improvement. If typed instructions produce that record, instructions inferred from brain signals need stricter checks, with a waking human approving anything that touches money, health or personal data.Fourth, mental privacy. Brain data is the most intimate data we have. In November 2025 UNESCO adopted the first global standard on the ethics of neurotechnology, and Chile amended its constitution in 2021 to protect brain activity. India’s Digital Personal Data Protection Act, 2023 does not treat neural data as a distinct category, a gap to close before such devices become everyday products. The Stanford team offered a useful design lesson: their decoder stayed locked until the user silently thought a chosen password, which it recognised more than 98% of the time.
Share your thoughts in the comments
Be respectful · TOI community guidelines
Consent was built into the machine.Finally, remember whom this technology serves first. These implants were built to give people with paralysis their voice and independence back, and that purpose must keep its priority.India has one of the world’s largest communities of software developers, and its institutes will help decide how these systems are built. My advice is simple. Treat signals from the brain as suggestions, never as commands. Keep a conscious, accountable human between intent and execution. Verify everything. Protect neural data in law. Keep teaching students to reason about software themselves. If we do that, we may one day wake up to an app that began as a dream and still be able to say that we understand it, we approved it and we answer for it.
Further reading
- S. Dey, “Neural Coding as Software Engineering Augmentation, Not Abdication,” Communications of the ACM, vol. 69, no. 8, pp. 32-35, August 2026. doi.org/10.1145/3779310
- S. Dey, “Why Neural Coding Is a Cognitive Trap as an Evolutionary Step From Vibe Coding
,” Computer (IEEE), vol. 59, no. 7, pp. 100-104, July 2026. doi.org/10.1109/MC.2026.3667294
Disclaimer: Views expressed above are the author’s own.
End of Article
Follow Us On Social Media
Photostories
Hot Picks
New Orleans cruise ship collision
