{"slug": "ai-is-not-a-source-of-truth", "title": "AI Is Not a Source of Truth", "summary": "A developer argues that AI coding agents should not be treated as sources of truth, emphasizing that their findings require verification through direct evidence such as running code or writing tests. The developer suggests that while agents are useful for identifying potential issues, their claims must be proven independently of the model's confidence.", "body_md": "Coding agents are pretty good at finding suspicious code.\n\nGive one a repository and it will happily spend an hour following calls, reading tests, checking configuration and coming back with a list of things that look wrong.\n\nThe awkward part starts after that.\n\nSuppose the agent tells me that a particular code path can leave the application in an invalid state.\n\nWhat do I actually know at this point?\n\nNot much.\n\nI know that the model found something worth checking.\n\nIt could be a real bug. It could also have missed a guard somewhere else, misunderstood how the code is called, assumed the wrong configuration, or simply produced a convincing explanation for something that cannot happen.\n\nI've become increasingly uncomfortable with treating those two things — finding a possible problem and proving the problem exists — as the same job.\n\nAsking another model isn't a great answer either.\n\nIf Claude finds a bug and another model agrees with Claude, that's useful. I probably want to investigate it sooner.\n\nBut agreement is still not evidence.\n\nIf possible, I want the claim to survive something that doesn't care what either model thinks.\n\nRun the code. Reproduce the behaviour. Write a test that fails for the reason being claimed. Check the actual configuration. Look at the log, the dependency version, the Git history, whatever can answer the question directly.\n\nSometimes that is easy.\n\nIf an agent says a function throws on a particular input, call it with that input.\n\nSometimes it isn't. Architecture and concurrency bugs are obvious examples. You may only be able to establish part of the claim, or discover that reproducing it would require assumptions you cannot confirm.\n\nThat's fine too.\n\n\"Couldn't verify this\" is useful information.\n\nI'd much rather keep a suspicious claim unresolved than turn it into a finding because a model sounded confident.\n\nThis is also why I don't think static analysis and coding agents are competing approaches.\n\nStatic tools are great when we already know what to look for and can describe it precisely.\n\nThe interesting thing about an agent is that it can notice something nobody thought to write a rule for.\n\nThat is the part I want from it.\n\nLet it dig through an unfamiliar repository and come back with weird questions.\n\nThen make those questions earn their way into the bug list.\n\nThe more code agents can write and inspect, the more this distinction matters. Generating another thousand lines is getting very cheap. Carefully checking whether a subtle claim about those thousand lines is true is not.\n\nSo when an agent tells me it found a bug, I don't want the next question to be:\n\n\"How confident are you?\"\n\nI want it to be:\n\n\"How can we prove it?\"", "url": "https://wpnews.pro/news/ai-is-not-a-source-of-truth", "canonical_source": "https://dev.to/rud0lforrp/ai-is-not-a-source-of-truth-48kc", "published_at": "2026-08-17 03:50:37+00:00", "updated_at": "2026-08-17 04:42:36.113418+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "developer-tools"], "entities": ["Claude"], "alternates": {"html": "https://wpnews.pro/news/ai-is-not-a-source-of-truth", "markdown": "https://wpnews.pro/news/ai-is-not-a-source-of-truth.md", "text": "https://wpnews.pro/news/ai-is-not-a-source-of-truth.txt", "jsonld": "https://wpnews.pro/news/ai-is-not-a-source-of-truth.jsonld"}}