There is a fight underway about the speed of artificial intelligence, and like most fights that generate more heat than light, it has hardened into two tribes who mostly talk past each other.
On one side stands accelerationism — in its current technological form, "effective accelerationism," or e/acc. Its claim is close to a physics of progress: technology, and AI above all, is a fundamentally good and near-inevitable force, and the moral error is braking it. Don't slow down; take the brakes off. The instinct has serious voices behind it — the spirit of Marc Andreessen's Techno-Optimist Manifesto, the founder-culture gospel of blitzscaling — and it defines itself against caution, which it tends to read as cowardice.
On the other side stands the deceleration instinct — the safety researchers and writers who look at the same curve and see risk outrunning wisdom, who ask for pauses, guardrails, and time. Its serious voices are serious too; think of Ian Hogarth's widely read warning about racing toward "God-like AI." This tribe reads acceleration as recklessness, sometimes as something close to a death wish.
You are invited, constantly, to pick a side. Speed or safety. Build or fear. We decline the invitation — not out of centrist timidity, but because the binary is measuring the wrong thing.
The race is being measured on the wrong axis
The argument treats velocity as the variable that matters — as if the only question were how fast the car should go. But speed was never the real defect. Plenty of fast things are honest, and plenty of slow things are rotten. The actual failure of the current race is what it quietly sacrifices in order to be fast.
Under competitive pressure, the first things cut are always the ones that are invisible in a demo and expensive to do right: where the training data came from and whether the people who made it agreed; whether a confident-sounding claim is actually true or merely fluent; whether a human stayed the author of the thought or was quietly replaced by it; who absorbs the harm when the system is wrong. None of these show up in a launch video. All of them are the difference between a tool that extends people and one that extracts from them.
Accelerationism's real sin is not that it moves fast. It is that it moralizes the trade — it reframes "we skipped the receipts" as boldness and "show me your sources" as timidity. It takes the erasure of the record and dresses it as courage. That is the rot in the race: not motion, but motion that erases provenance while calling erasure a virtue.
The acceleration ditch
Speed as its own justification. Caution is weakness; provenance and consent are friction to remove. Externalized risk, reframed as boldness.
The deceleration ditch
Safety as its own justification. Progress is suspect; the answer is to slow, pause, gatekeep — a posture incumbents are happy to capture and weaponize.
The honest concession
Here is where we part from our own tribe's temptation, because a position that can't survive the other side's best punch isn't worth holding. The accelerationists are not wrong about everything. Stagnation has a body count too — diseases left uncured, poverty left unrelieved, problems left unsolved because we lost our nerve. And "safetyism" genuinely does get captured: incumbents learn to love the regulation that happens to freeze the frontier exactly where they already sit, pulling the ladder up behind them in the name of caution. Anyone who tells you the slow road is simply the good road is selling something too.
So we don't want the race to stop. We are builders; we think the vision is worth reaching. We just refuse to reach it by lying.
The third road: acceleration with provenance
The way out of the binary is to notice that the thing being sacrificed — honesty — is not actually the same variable as speed. You can move quickly and carry the record, if you build the record in from the beginning instead of bolting caution on after the fact. Honesty doesn't require slowing the car. It requires keeping the ledger while you drive.
This is not an abstract hope for us; it is an old craft. Interveil's founder spent two decades in forensic audio, in the lineage of the tools that restored damaged film sound — pulling signal from noise, in real time, without falsifying the record. That is the whole ethic, and it turns out to scale from a soundtrack to a civilization. Restore the signal fast. Never fake what wasn't there. Where the source is missing, say so rather than confabulate.
Concretely, that means an AI posture with a few non-negotiables: every claim carries its evidence class and its source, so fluency can never impersonate truth; the human stays the author, with the machine priming and whispering but never quietly deciding; enrichment is offered with consent and a receipt, never harvested in the dark; and autonomy is deny-by-default, earned rather than assumed. None of these slow the work in any way that matters. They simply make it impossible to win by cheating — which is the only kind of speed worth refusing.
Between veil and vision
Our emblem line is "Between Veil and Vision." The veil is the membrane between a human mind and everything mediated; the vision is the future worth building on the other side of it. The two tribes each grab one and drop the other — the accelerationists chase the vision and tear the veil; the decelerationists guard the veil and lose the vision. We are trying to hold both: to keep moving toward what AI can genuinely give us, while keeping the membrane honest the entire way.
We are not asking the race to slow down. We are asking it to stop lying. Those were never the same request — and almost the whole argument has been conducted as if they were.
Interveil Labs makes human-agency-preserving AI for media and mind. This essay states Interveil's position and argues for it; it presents the opposing views as their strongest advocates would. Drafted with AI tools, edited and stood behind by D. Hardwick. We make the veil honest.
Sources & further reading
- Reid Hoffman & Chris Yeh, Blitzscaling (2018) — the speed-over-efficiency growth doctrine, and its explicit conditions.
- Marc Andreessen, “The Techno-Optimist Manifesto” (a16z, 2023). a16z.com
- On effective accelerationism (e/acc): origins and framing, 2022–2023 (Verdon / “Beff Jezos” and others).
- Ian Hogarth, “We must slow down the race to God-like AI,” Financial Times (2023) — the deceleration pole.
- NIST AI Risk Management Framework — accountability and documentation as practice, not slogan. nist.gov