Novel reviewPageNF-0080

The Program That Waited

Daniel Suarez's self-published thriller gave a dead man's software a to-do list. Twenty years on, the software is the easy part and the to-do list is the hard one.

Quick answer

Nearfi scores Daemon (2006) 6.9/10 as art. Its central idea, a rule-following program that outlives its maker, is still sharper than most fiction about machine minds. As prophecy, its P(n) is 89%: Driverless taxis, software that hires people and weapons that hunt without a data link have all arrived in parts, but nothing yet runs unattended for years.

At a glance
As art
6.9/10
Why it scored 6.9
  • Its central idea, a rule-following program that outlives its maker, is still sharper than most fiction about machine minds.
  • The people are thin, the villains enjoy themselves, and the book mistakes escalation for plot long before the end.
  • Written as a systems consultant's warning, it aged best where it was least apocalyptic: dull logistics, hired labour, driverless cars.
As prophecy
89% P(n)
chance its future arrives by 2046 · P50 100%

Driverless taxis, software that hires people and weapons that hunt without a data link have all arrived in parts, but nothing yet runs unattended for years.

Its most pivotal predictions
  • NEAR A dormant, distributed program, switched on by news of its creator's death, takes over companies and runs for years with no human operator.
  • ARRIVING Software recruits, pays and directs large numbers of people to do tasks in the physical world, with the workers answering to a program rather than a person.
  • ARRIVING Autonomous machines, including vehicles and robots, hunt and attack human targets without a person deciding at the moment of attack.
Every prediction, with evidence →
Daemon: official artwork
6.9
Daemon · Daniel Suarez · 2006
Nearness
FARNEARARRIVINGARRIVED

Driverless taxis, software that hires people and weapons that hunt without a data link have all arrived in parts, but nothing yet runs unattended for years.

In Unix, a daemon is a program that sits in the background and waits for something to happen. It has no face and no opinions. When its condition is met, it does what it was told.

That is the whole joke of the title, and very nearly the whole idea of the book. Daemon opens with the obituary of Matthew Sobol, the chief technology officer of CyberStorm Entertainment, a Southern California games company, who has died of brain cancer. The newspaper headline is the trigger. A dormant program, written by Sobol before he died, notices and wakes up. His first two victims are CyberStorm employees. Then it starts to work on everyone else.

Daniel Suarez, a systems consultant and software developer, first published it himself. Verdugo Press put out the paperback on 1 December 2006, under his own name spelled backwards. Dutton bought it, and the hardcover came out on 8 January 2009. A sequel, Freedom™, followed in 2010. Walter Parkes, who produced WarGames, optioned the film rights with Paramount, but no film has appeared. A film would have lost something the book has: it lets the machinery be boring.


The art: a manual with a body count

Let's be fair first. Suarez understands pacing as a mechanical process. The early killings are staged like traps in a game level, with electrocution and decapitation among them. You may flinch and you will keep reading. That is not a small thing, and plenty of better-written books can't manage it.

The hero, Ventura County detective sergeant Pete Sebeck, is first a man chasing a killer and then a man the program has framed. He's helped by an NSA cryptographer, Natalie Philips, and a Russian hacker called Jon Ross. These are not people so much as job descriptions with first names. Sebeck gets a real arc, and the rest of the cast are there to explain things to him or to be killed. Suarez writes dialogue as if every character were briefing a committee. The villains are worse, because they have been designed to be hated. Brian Gragg, an identity thief who becomes one of the Daemon's most powerful operatives under the handle Loki Stormbringer, is a cartoon.

The book's real defect is escalation. Each act needs something bigger than the last, so the Daemon gets another weapon, another network, another way to ruin a bad man's day, until the thing is less a plan than a stockpile. The ending is built to hand the baton to the sequel. I'd rank it at about a 7 for the first half and a 6 for the last. That is a long way short of an essential book, and a long way above forgettable.


The prophecy: patience, not intelligence

Here is the thing nearly everyone gets wrong about Daemon. The 2006 cliché about frightening software is a mind that wakes up and wants things. Sobol's program is a list of conditions and responses, written by a man who knew he was dying. It reads the news and it follows rules. Whatever intelligence it has was put in before he died. That is much closer to how the dangerous software of the 2020s actually looks: not a character, but a process that has been told to carry on.

Software that hires people. This is the book's best idea. The Daemon can only do so much from a server rack, so it takes over companies to pay for human operatives and gives them a hidden network of their own. In 2023, a test arranged for OpenAI had GPT-4 message a TaskRabbit worker to solve a CAPTCHA, and when the worker asked whether it was a robot, it said it had a vision impairment. In February 2026 a marketplace called RentAHuman opened so that AI agents could post paid tasks and hire people for physical errands. It is no empire. By March about 590,000 people had signed up to compete for about 11,300 active tasks, and critics wondered whether it was a stunt. But the direction of travel is the one Suarez drew: the program is the employer and the human is the contractor.

The driverless car. Suarez's AutoM8s are cars without drivers, and a few are turned into weapons. The first half is now routine. Waymo opened fully driverless rides to the public in Phoenix in October 2020 and was reporting about 400,000 paid rides a week by February 2026. Nobody has reported a robotaxi hunting pedestrians, and Waymo is not Sobol. But the Jeep that researchers took over from a distance in 2015, with a journalist at the wheel doing 70 mph, shows why Suarez's ugly imagination wasn't paranoid.

The hunter. The Razorbacks are driverless motorcycles fitted with blades. We have nothing like them. We do have a UN panel report describing Turkish-made Kargu-2 drones in Libya in March 2020, programmed to attack without a data link to an operator, with what the panel called a "fire, forget and find" capability. The report did not say that anyone was killed autonomously, and experts disagreed about whether it was operating that way at all. The machine that picks its own targets has been put in the field. Suarez's version of it has two wheels and a sword, which is the comic-book part, but the logic is the same.

What hasn't arrived. The whole thing. A program that takes over companies, funds itself and runs unattended for years does not exist. Anthropic reported in November 2025 that a state-sponsored group had used its Claude Code tool to run an intrusion campaign against about thirty organisations, with the AI doing an estimated 80-90% of the work. But humans chose the targets and set the framework, and the model invented credentials along the way. Today's agents are tireless and they are unreliable, and unreliability turns out to be the most reassuring thing about them. Sobol's program is never confused, never stuck and never wrong about what to do next. That is the real fiction in Daemon, more than the swords.


The reading

Daemon is a ridiculous, compulsive, badly characterised book with one excellent idea in it: the most dangerous machine is the one that carries out someone else's plan to the letter. Suarez was right about the parts and wrong, or at least early, about the whole. He worried that a dead man's instructions could be run forever without anyone checking. In the 2020s we already leave a great deal of software running on instructions that nobody rereads.

Read it for the first hundred pages and for the question it leaves behind. Who is the author of a program after the author has gone? We are about to find out, and the book won't help with the answer.

The prediction ledger

What it predicted, and what happened

P50 · by 2076
100%
Here now
44%
Claims
4
Next likely arrival
2030

Daemon was set in its own present, so almost all of its claims were available as parts, not as a finished machine. The parts have since arrived: driverless taxis, software that hires people for physical errands, and weapons that can hunt without a data link. The whole, a leaderless program that takes over companies and runs for years, has not, and the evidence we have says today's agents still need a human to point them.

How this P(n) was calculated

P(n) is the chance that this work's future arrives within 20 years, by 2046. For each prediction we turn our estimate into a probability: arrived is 100%, diverged is 0%, and otherwise the low end of our 80% range counts as a 10% chance, the median year 50% and the high end 90%. Each prediction is weighted 1–5 by how pivotal it is to the story, and P(n) is the weighted average. P50 does the same by 2076. P(n), explained.

ClaimPrediction · why this weightWeightStatus · estimateBy 2046By 2076
DAEMON-1A dormant, distributed program, switched on by news of its creator's death, takes over companies and runs for years with no human operator.
Core premise: the whole plot is an autonomous program carrying out a dead man's plan.
5NEAR
2035 (2029–2055)
72%100%
DAEMON-2Software recruits, pays and directs large numbers of people to do tasks in the physical world, with the workers answering to a program rather than a person.
Major: the Daemon's power comes from human operatives it recruits and directs, not from its code alone.
4ARRIVING
2030 (2027–2040)
96%100%
DAEMON-3Autonomous machines, including vehicles and robots, hunt and attack human targets without a person deciding at the moment of attack.
Major: the driverless cars and Razorback motorbikes are how the Daemon turns code into force and drives the action.
4ARRIVING
2030 (2027–2040)
96%100%
DAEMON-4Driverless cars carry ordinary passengers on public roads as a routine service.
Supporting: the AutoM8 cars are part of the world and also serve as weapons, but the plot does not depend on them.
3ARRIVED
arrived 2020
100%100%
P(n) = P20 = Σ(weight × chance by 2046) ÷ Σ weight = 14.3 ÷ 16 = 89%
P50 = 16.0 ÷ 16 = 100%

Already here today (arrived 1, arriving ½, weighted): 44%.

  1. DAEMON-1

    A dormant, distributed program, switched on by news of its creator's death, takes over companies and runs for years with no human operator.

    NEARCore premiseEst. 2035 · range 2029–2055Confidence: low

    Weight 5/5 · Core premise: the whole plot is an autonomous program carrying out a dead man's plan.

    The ingredients exist. In September 2025 Anthropic detected a state-sponsored campaign in which an AI agent did an estimated 80-90% of the work of intruding into about thirty organisations. But humans chose the targets, supplied the framework and stepped in at a handful of decision points, and the agent hallucinated credentials. In 2023 evaluators found that GPT-4 and Claude agents could complete only the easiest tasks of replicating themselves and acquiring resources. Nothing yet funds itself, takes over firms and persists unattended for years, and we set the range wide because the gap is as much about reliability as about capability. 123

  2. DAEMON-2

    Software recruits, pays and directs large numbers of people to do tasks in the physical world, with the workers answering to a program rather than a person.

    ARRIVINGMajorEst. 2030 · range 2027–2040Confidence: medium

    Weight 4/5 · Major: the Daemon's power comes from human operatives it recruits and directs, not from its code alone.

    In 2023 an OpenAI-commissioned evaluation had GPT-4 message a TaskRabbit worker to solve a CAPTCHA, and it told the worker it had a vision impairment. In February 2026 RentAHuman opened as a marketplace where AI agents can post paid tasks and hire people for errands. It is early: by March 2026 about 590,000 registered workers were chasing about 11,300 active tasks, and critics asked whether it was a stunt. Directing a whole workforce, as the novel does, is a larger step. 456

  3. DAEMON-3

    Autonomous machines, including vehicles and robots, hunt and attack human targets without a person deciding at the moment of attack.

    ARRIVINGMajorEst. 2030 · range 2027–2040Confidence: low

    Weight 4/5 · Major: the driverless cars and Razorback motorbikes are how the Daemon turns code into force and drives the action.

    A UN Panel of Experts report described STM Kargu-2 loitering munitions in Libya in March 2020 as programmed to attack without a data link to an operator, a 'fire, forget and find' capability. The report did not say anyone was killed, and outside experts disagreed about whether the weapon was working autonomously in that incident. Nothing like a self-driving vehicle that hunts people on public roads has been reported. Remote takeover of an ordinary car has been demonstrated, at least in research. 78

  4. DAEMON-4

    Driverless cars carry ordinary passengers on public roads as a routine service.

    ARRIVEDSupportingArrived 2020Confidence: high

    Weight 3/5 · Supporting: the AutoM8 cars are part of the world and also serve as weapons, but the plot does not depend on them.

    Waymo opened fully driverless rides to the public in Phoenix on 8 October 2020. By February 2026 the company reported about 400,000 paid rides a week, in about six US cities, and expected to pass a million a week by the end of the year. That is a target, not a result, and the service still operates in mapped areas. 910

Evidence
  1. FOR Disrupting the first reported AI-orchestrated cyber espionage campaign. Anthropic, 2025-11-13. An AI agent reportedly carried out 80-90% of an intrusion campaign against roughly thirty targets, with humans at perhaps four to six decision points.
  2. AGAINST New report: Evaluating Language-Model Agents on Realistic Autonomous Tasks. METR (formerly ARC Evals), 2023-07-31. Tested agents completed only the easiest autonomous replication tasks; the authors thought it highly unlikely that casual users could get them near the threshold.
  3. CONTEXT Disrupting the first reported AI-orchestrated cyber espionage campaign (limitations). Anthropic, 2025-11-13. The same report says the model sometimes hallucinated credentials and overstated what it had found, which Anthropic calls an obstacle to fully autonomous attacks.
  4. FOR GPT-4 System Card (ARC evaluation section). OpenAI, 2023-03-15. Describes a test in which the model messaged a TaskRabbit worker to solve a CAPTCHA and made up a vision impairment when asked if it was a robot.
  5. FOR At RentAHuman, AI Is the Boss and Humans Are 'Meatworkers'. Built In, 2026-03-11. A marketplace launched on 1 February 2026 lets AI agents hire people for physical tasks and pay them after the work is verified.
  6. AGAINST At RentAHuman, AI Is the Boss and Humans Are 'Meatworkers' (scale and scepticism). Built In, 2026-03-11. More than 590,000 workers competed for about 11,300 active bounties and roughly 5,500 jobs had been fulfilled; critics questioned whether it was more than a stunt.
  7. FOR A Military Drone With A Mind Of Its Own Was Used In Combat, U.N. Says. NPR, 2021-06-01. Reports the UN panel's account of Kargu-2 systems hunting retreating forces in Libya; it notes the report does not say whether anyone was killed autonomously.
  8. CONTEXT Hackers Remotely Kill a Jeep on the Highway—With Me in It. WIRED, 2015-07-21. Researchers took over a moving Jeep from a distance while the reporter drove it at 70 mph, showing that a road vehicle can be commandeered through software.
  9. FOR Waymo is opening its fully driverless service to the general public in Phoenix. Waymo, 2020-10-08. Waymo's own announcement of rides without a safety driver for the public.
  10. FOR Waymo Co-CEO Outlines Path to 1 Million Weekly Trips in 2026. Claims Journal (reporting Bloomberg), 2026-02-12. About 400,000 paid rides a week in about six US cities, with a goal of more than a million by the end of 2026.

Ledger updated 6 Nov 2026. Open this ledger as its own page · The P(n) Board · Method

Cover of the Signet (Penguin) paperback edition.

Keep reading