Two hundred and fifty years ago, 56 guys got together and signed a decision record.
We remember the Declaration of Independence as rhetoric, and it’s very good for that. But turn down the patriotic hymns and listen to the document for a moment.
- There is context: “when in the course of human events.”
- There is evidence for inspection: “let facts be submitted to a candid world, followed by twenty-seven enumerated grievances.”
- There is a decision: “we therefore publish and declare.”
And then there’s the part we probably quote the least: the named people who signed it. They each staked their most cherished values on being wrong, and meant it: “our lives, our fortunes, and our sacred honor.”
No runtime. No service. Just text. Yet it outlived every single institution it addressed. (Well, I guess there’s still a monarchy, but the nature of it certainly ain’t what it used to be.)
One more detail from that famous week in Philadelphia, though slightly less famous (especially these days, unfortunately). The draft that reached Congress contained a grievance condemning the slave trade. Congress struck it. We know exactly what was cut, and who objected to the cutting, mostly because Thomas Jefferson kept his own copies and spent years showing people exactly what the record had lost. I was fascinated with this stricken language when I visited his home in Monticello years ago, because I had no idea. Apparently, the dissent survived by luck and our Forefather’s bruised pride. But that’s not a very good preservation strategy. What gets overruled isn’t just part of history, it’s actually part of the decision itself. It shows intent, it foreshadows the future, and history needs more than simply one guy’s scrapbook to know that it happened.
Anniversaries are checkpoints. Today, as we celebrate America’s 250th, to me the question worth asking isn’t about nations. It’s about what independence is worth our declaring it right now. Here we are, in the middle of the fastest pace of change human history, deciding in how decisions get made that any of us will manage to live through. Let alone our children, or another 10 generations to follow.
I do a lot of very different things, but one of them is building open protocols for how humans and AI systems decide things together. From inside that work, here is what I think 2026 needs independence from.
Independence from the smartest mind in the room
Nearly everything built in The Age of Agents assumes intelligence is a property of the best individual AI model. Benchmark it, buy it, route your hardest problem to it. Okay… but the most consequential intelligences in the world have never lived in a single mind, human or otherwise. It’s always a composite. A surgeon and a team. A captain and a crew. I’ve written dozens of examples like this in my weekly newsletter Signals & Subtractions, also canonically represented on this very blog. The performance that matters is the performance of the whole, and the whole is never the average of the parts. Depending on how those parts are arranged, we get different results. It depends on how they hand off, how they catch each other’s failures, how they disagree.
The unit that matters is not the thing. It is the relationship between the things. “More edge than node” as I’ve come to find myself saying a lot lately.
Independence from the boss model
Just over 100 years ago, scientific management separated planning from execution. Thinking moved to the planning office, and workers became hands. It built the modern world’s throughput, and that’s still being rebuilt right now, almost verbatim, as agent orchestration: one premium model writes the plan, cheap workers execute in parallel, and verification checks whether the work ran.
For labor, this is arguably correct. But judgment is not really labor, is it? One mind with many hands is just one blind spot with a force multiplier. Verification can prove that every task passed, cool. But nothing in the architecture can prove that the plan ever deserved to exist at any point in its existence. The failure of the agentic era will not be work done badly. It’ll be the wrong work done flawlessly, quickly, and at unbelievable scales.
Independence from consensus
A new class of tools now runs your question past several models at once. But just watch what most of them do with disagreement. They surface it, then dissolve it down into one confident, synthesized answer. The friction, which was the most informative (and important!) part, disintegrates on contact.
Agreement is evidence, sure, but that’s not the same thing as authority. When three models converge, that’s worth knowing. When they diverge, it’s worth even more, and a system that morphs divergence into smoothness isn’t finding certainty as much as manufacturing it.
We people do this, we call it groupthink. We don’t generally use that word to mean good things, but we’re now automating it and making it go faster than humans ever could.
Independence from the vendor leash
If one company owns the layer where all the groupthinky collaboration norms live, then the norms are whatever serves that company and their shareholder’s interests. Not the customers, we’ll just get what they ask for on our behalf (if they remember to do that at all). I believe that the rules for how minds work together (both human and machine) have to be a commons: open specifications, plain files, no gatekeeper service, verifiable by anyone from the artifact alone. Because a standard you can’t inspect and can’t ever leave is not a much of a standard. It’s more like a leash with documentation. No thank you, King George, or whatever name you go by these days.
The part I couldn’t write last 4th of July (or any before that)
This year, 2026, independence finally became a conformance claim.
In a decision-record format that I maintain, a record earns a claim called independent when it shows two or more positions from distinct model families, recorded before any human ruled on them. A zero-dependency script checks it. Anyone can run the check, from the file alone, with nothing but Node and the record itself.
Those mechanics matter to me. Positions are gathered in isolation because the first voice tends to shape the room. Every facilitator has watched it happen to a group of otherwise very smart people. Distinct model families are required because a monoculture cannot produce honest dissent, it only echoes with slight variances. Independence, in this narrow and useful sense, stopped being rhetoric two hundred and fifty years after Philadelphia and became something you can verify with a shell command. I think that’s pretty cool. Foundational, even.
Declarations are the easy part
Here’s what today’s anniversary actually teaches, if you read the history like I do. Independence was declared in 1776, and then it nearly failed for eleven years straight, because declaring independence turned out to be much simpler than making it work in practice. The working system finally arrived in 1787, when most of the same guys sat back down and did the unglamorous and always contested work of arranging a balance of powers:
- who proposes
- who checks
- who decides
- who answers for it
That same order of operations applies now. Independent positions are raw material. The intelligence worth having comes from the arrangement built on top: dissent recorded rather than smoothed, objections preserved rather than deleted, and one accountable human writing the decision. Not because humans are smarter or better. We are at some things, I guess, but we’re slower and we are jagged with our intelligence too. The real reason for keeping accountability in the people lane is that this has to land on someone who lives with the consequences, and a synthesized answer can’t offer up “our lives, our fortunes, and our sacred honor” from agents.
AI models can hold positions, but a person holds the pen.
I maintain two open standards built on exactly this: Turnfile, a governance protocol for sessions where peer agents disagree before acting, and AIDR, a one-file decision record for the moments that matter. Both are plain text. Both are free. Both assume disagreement is the engine, not the exhaust. I’m working on a third piece that builds on top of both, more news on that soon.
I’m building any of it because I can clearly see that there is only a narrow window in which humans and machines will still read each other’s reasoning, and every legible record we write is practice for keeping it open.
250 years on, the most durable thing about that sacred document we celebrate is not the prose. It is that the people who decided put their names under the decision, in plain text, where anyone could read them, forever.
It is worth celebrating. And now it’s worth verifying too.
P.S. I may be a little more biased now that my genealogy research recently showed that one of those names on the Declaration of Independence was signed by a certain Pennsylvania politician that was my great, great, great, great, great, great, great Grandfather (on my Father’s side). How fitting.