Jim Lau is selling confidence to people who need permission to feel smart again.
Flowing is his desktop application for researchers and students who want to write with AI assistance without hallucinating their way into a failed dissertation. The tool anchors every claim to a source — you cite as you compose. The algorithm stays tethered to the documents you've already read.
In theory, this solves the hallucination problem. In practice, it replicates a disaster that happened thirty years ago in law libraries, unfolding in real time because the mechanism is so elegant we don't recognize it as dangerous.
In 1995, legal databases like LexisNexis made full-text search across millions of court opinions available for the first time. A corporate lawyer arguing that precedent supported nearly any position could query the database, find a citation. Cite it accurately. The problem wasn't that the citation was wrong — it was that it became trivially easy to construct an argument from selective quotation when you had access to enough volume. The sheer existence of a citation, accurate and properly formatted, became proof of diligence. Nobody asked whether the precedent actually supported the argument.
You're no longer reasoning toward evidence. You're reasoning within a haystack and calling the needle you found vindication.
”Evidence-grounding in AI writing creates the same structure of composing a claim, having the algorithm find a real source to support it. Landing that citation in your text with confidence. The source is real, the quotation is accurate. But what disappears is the act of thinking about whether that particular source actually supports that particular claim. You're no longer reasoning toward evidence. Instead, you're reasoning within a haystack and calling the needle you found vindication. The difference that might change this ending isn't technological, it's social. LexisNexis created a monoculture of argumentation because lawyers had no choice. Flowing exists in a world where alternatives still matter. If enough readers and advisors start asking not "Is this cited?" but "Is this citation doing actual work here?" the tool becomes useless, and you're back to thinking.