The context was in the meeting
a16z published an essay recently called You Just Hired a Million Bad Employees. The short version: for the first time, humans are cheaper than software. Every employee now has effectively unlimited headcount through AI, and most of it is wasted. Tokens get burned without direction. Agents loop on themselves, spending tokens on spending tokens. The scarce skill, the essay argues, is management: converting fuzzy intent into instructions a machine can actually execute.
Two lines in it stopped us. First, that loops are not sophistication, they are a symptom: they happen because "humans never articulated the task cleanly." Second, that "maybe 1 in 100 employees knows how to give AI context."
We build a meeting recorder, so we have a view on where that context lives.
The task was articulated cleanly, once
Here is the thing the essay circles but never quite lands on: almost every task does get articulated cleanly. It happens out loud, in the meeting where the work was decided. Someone asks the naive question, someone names the constraint nobody had written down, someone explains why the last approach failed, and the room converges on what the work actually is and why.
That version, the clear one, exists for about an hour. Then everyone leaves with a two-line action item. Three weeks later an agent gets a two-line prompt, which is the fuzzy residue of a conversation nobody kept, and it loops.
The essay's second imperative is "record what works." We would move one step upstream: record what was said. The articulation problem is not that people can't state tasks clearly. It's that the clear statement is spoken, not written, and spoken words evaporate.
A folder of Markdown is a context layer
Boswell records and transcribes meetings on your Mac and files them as plain Markdown. The format is not incidental. Markdown is the one thing every model, every agent, and every tool you will use this year already reads. A folder of dated meeting notes is a context layer with no platform, no integration, and no vendor between you and it. You point a tool at a file, or paste the three paragraphs that matter.
The essay calls the efficient version "100X tokens," prompts that do orders of magnitude more per token spent. In our experience the difference between a looping agent and a one-shot answer is rarely a cleverer prompt. It is the background the prompt was missing: the decision from the kickoff, the constraint the customer named on the call, the reason the last approach was abandoned. All of that was said in a meeting. The cheapest 100X token is a paragraph from notes you already have.
The hoarding problem looks different from here
The essay's darkest observation is "context hoarding": employees withholding what they know from AI systems because nobody trains their replacement for free. It cites the backlash at Meta when employees realized their work context was training data.
Fair. But notice what the framing assumes: that giving AI context means pouring your working knowledge into a system your employer owns. Of course people balk at that.
Boswell's model is a different deal. The notes belong to the person who took them, plain Markdown files with no platform between you and what you know. Your accumulated context stays a personal asset. You decide which file goes to which tool, and whether it goes at all. People have always kept notes for themselves. Nobody has ever needed an incentive plan for that.
What we don't claim
Boswell will not write your evals, restructure your org, or tell your agents what to do. The essay is right that those are management problems, and no app solves them. We do one narrow thing: we keep the moment when a task was articulated cleanly from evaporating. It happens to be the input everything else in the essay depends on.
The record outlives the talk
The essay closes: "Humans just became cheaper than software. Someone still has to tell them both what to do." The telling mostly happens out loud, in rooms and on calls. Our namesake built a career on writing down what a brilliant talker actually said, and two centuries later it is the record, not the talk, that survives. That is the job. The context was in the meeting. Keep it.