The latest articles on DEV Community by Panu Horsmalahti (@panuhorsmalahti). https://dev.to/panuhorsmalahti https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4103717%2Fb96d0f67-5f0d-45c5-b8ba-f60375ac90d8.jpg https://dev.to/panuhorsmalahti en Panu Horsmalahti Tue, 01 Sep 2026 05:55:43 +0000 https://dev.to/panuhorsmalahti/will-software-ever-be-free-209a https://dev.to/panuhorsmalahti/will-software-ever-be-free-209a <p>Lets assume that in the near future the following will be true:</p> <ul> <li>Coding agents are better than 99% of software engineers</li> <li>Tokens are practically free</li> </ul> <p>If and when we get there, will software be "solved"? Will code become free to produce at a high quality? My answer is.. not quite, and the reason is data.</p> <p>I don't just mean "big data", the data that large companies utilise to steer their services. I also mean "small data". Small data are the large number of small decision one has to make when creating software. This is called the "spec".</p> <p>LLMs can of course make these decisions. And once models get better, they ought to get better at it. However, I don't think training data will include all the possible decisions for all software. Some spec will remain essentially proprietary. This means humans will have to make the choices, and this will take time and effort.</p> <p>This time and effort will therefore impose a lower bound for the price of software. Companies will of course find the right answers over time, but the price of software will remain non-free.</p> ai llm software