The Phoenix Project: A Novel About IT, DevOps, and Helping Your Business Win

Miles: It's a Tuesday morning at a car parts company, and the hourly workers are not going to get paid. The payroll run has failed, and only for them. Everybody on salary came through fine. Everybody paid by the hour has a blank where a number belongs.

Miles: Nobody can say what changed. Plenty changed. Nobody wrote any of it down.

Miles: The man who has to fix it was promoted into the job about four hours earlier, and he'd turned it down twice.

Miles: Today: The Phoenix Project, by Gene Kim, Kevin Behr and George Spafford. In two sentences: a technology department is a factory, it has one slowest station like every other factory, and most of the damage there comes from work nobody planned and nobody counted. The repair is a cap on how many things can be in flight at once, and that cap is the closest thing in this series to an adjustment that arrives without anybody having to request it.

Miles: Sixty Books. Presented by Stoar Consulting and Coaching.

Miles: Book thirty two, and you're back in the reading room. This is Sixty Books, Miles speaking. The setup never moves: one management title an episode, sixty of the sturdiest of the past hundred years, each put to the same pair of tests. What's in it for a person who leads, or works under, a brain running on different wiring. And what survives a job that now has AI sitting in the middle of it. Nora holds the evidence, and she assembled this one out of it before either of us wrote a line.

Miles: Nora, we finished yesterday saying the factory floor had turned into the thing that ships code. This book does that literally.

Nora: With a car plant in it. The mentor drives the hero out to one of the company's own manufacturing sites, walks him along the catwalk above the floor, and tells him that if he thinks his department has nothing to learn from that room, he's badly wrong. The bridge you named gets drawn in ink.

Miles: And the correspondences run past the scenery.

Nora: They're deliberate, and the publisher says so plainly. The physicist who only ever asks questions becomes a prospective board member. The plant becomes a technology department. And Herbie, the slowest boy on that hike, becomes an engineer named Brent, who happens to be the one person at the company who knows how anything actually works.

Miles: Every organization has a Brent.

Nora: Yeah. Every organization has a Brent, and most of them have built a career ladder that quietly manufactures more of him.

Miles: Set the thing up.

Nora: Parts Unlimited makes car parts, it's losing badly to a competitor, and its answer is an enormous software programme called Phoenix, meant to put its retail business back in contention. Phoenix is late and over budget and has been for years. Bill Palmer runs a corner of technology operations and is handed the whole department the morning his two bosses are fired.

Miles: And the payroll failure from the cold open.

Nora: Lands in his first hours. It traces to a change nobody recorded, tangled with a hardware failure, and while he's inside that, the chief executive wants Phoenix shipped and the head of retail wants somebody named as the reason it isn't. The engine of the novel is that everything there is urgent and nobody can say what anybody is working on.

Miles: Enter the mentor.

Nora: Erik Reid, a candidate for the board, who behaves the way Jonah does in the other book. He won't answer a question. He asks two more and leaves. His opening demand is that before they discuss controlling work, Bill has to be able to say what work is.

Miles: Which sounds like a riddle.

Nora: It's the spine of the book. Bill can't answer, because his department has no list. Tickets, escalations, projects, somebody leaning over a desk at four in the afternoon: those arrive by different doors and never meet anywhere. You cannot manage a flow you have never once written down.

Miles: The authors, and there are three of them.

Nora: Gene Kim first. Purdue, where as an undergraduate in nineteen ninety two he wrote the original version of a security tool as an independent study project. He co founded the company that sold it in ninety eight and was its chief technology officer until twenty ten. The tool has a name, and we're not going to use it or rate it.

Miles: The second.

Nora: Kevin Behr, who founded a research body called the Information Technology Process Institute and has been a chief technology officer and a chief information officer at companies from large public ones down to startups. Third, George Spafford, a research director at Gartner covering exactly this ground, technical change and release management.

Miles: And they'd written together before.

Nora: On handbooks about making technology operations visible and auditable, which is where all three met this material.

Miles: And the debt to the other book is acknowledged.

Nora: Out loud. The publisher's account has Kim saying he and his co authors wanted a version of The Goal for technology for more than ten years, and that they sat down with Goldratt himself in two thousand four about applying constraint theory here. So it isn't an echo somebody spotted afterwards. It's a deliberate remake, discussed with the original author in the room.

Miles: And it landed.

Nora: January twenty thirteen. At the ten year mark the publisher put it at over seven hundred thousand copies and twelve languages, which is their figure rather than an audited one.

Miles: Now the machinery, in the book's order.

Nora: It starts with Erik's question. What is work. The novel's answer is four kinds, and a department that can't sort its day into them is flying blind. Business projects, what the company asked for. Internal projects, what the department needs to survive. Changes, meaning every adjustment anybody makes to anything. And unplanned work.

Miles: And the fourth one is the villain.

Nora: The fourth is the only one you didn't choose. The book's image for it is antimatter: planned work doesn't get politely delayed by unplanned work, it ignites on contact. And the sharp part is where it comes from. It isn't weather. It's the bill for changes made badly, or made twice, or made by somebody who told nobody.

Miles: So the first three generate the fourth.

Nora: Which makes it reducible rather than a fact of life. Firefighting stops being a personality and becomes an output with a source.

Miles: And then the constraint.

Nora: Brent. Every escalation of consequence routes through one engineer, because he's fast and willing and knows where the bodies are. The plant tour is what lets Bill see it. Brent is a work centre with a queue in front of it, and every improvement made anywhere else lengthens that queue.

Miles: So they build the fence.

Nora: They build a fence. Work reaches him through one door, somebody polices the door, side channels count as violations, and he isn't permitted to fix anything he hasn't been scheduled to fix. Then they sit people beside him to write down what he does as he does it, so the knowledge leaves his head. It reads as cruel and it's the kindest thing anybody does for him in the entire novel.

Miles: Mm-hm.

Miles: And the other two moves.

Nora: The change board, which is the least sophisticated technology in a book about automating everything. Index cards on a wall. Everybody posts what they intend to change and when, and two teams can suddenly see they're about to touch the same thing on the same night. Then the freeze. The chief executive agrees to stop starting new project work until the department finishes what's open.

Miles: That's a work in progress limit at company scale.

Nora: That's the whole book in one decision, and it's the one that costs a career to make. Nobody has ever been promoted for the projects they declined to begin.

Miles: Say more.

Nora: Erik gives Bill a piece of arithmetic for why the freeze works, and it's the direct inheritance from yesterday. Wait time, he says, is the percentage of time a resource is busy divided by the percentage of time it's idle. At half busy, a task waits about as long as it takes to do. At ninety percent busy, nine times as long. Same people, same skill, nine times the delay.

Miles: And we've met that term.

Nora: We ran it yesterday. Kingman's utilization factor from nineteen sixty one, utilization over one minus utilization, lifted out of a mathematics journal and put in a character's mouth fifty two years later. Flag this now, because it matters shortly: Kingman multiplies that factor by two others, and the book's version keeps one.

Miles: And the Three Ways.

Nora: The frame the whole DevOps movement took from here. First Way, flow, left to right, business through to customer, never passing a known defect downstream. Second Way, feedback, right to left, shortening the loop so problems return fast enough to be fixed by whoever caused them. Third Way, repetition and experiment, on the argument that mastery comes from practising the risky thing until it stops being risky.

Miles: And the constraint moves.

Nora: Goldratt's fifth step, on schedule. Once Brent isn't the limit, the limit walks outward, and by the end it isn't in the technology department at all. It's in the company's ability to say what it actually wants built. Hold that, because it's tomorrow.

Miles: Where does it stop?

Nora: At the same place the other one stopped. Start with the chart, because it's the most checkable claim in the book. Busy over idle is the utilization term of a queueing approximation, and it's genuinely right about the runaway. What it leaves out is the variability term, built from how much arrival times and job times vary. At identical utilization, a system with high variation queues worse than one with low variation.

Miles: And that matters here because.

Nora: Because knowledge work is the high variation case. Two tasks that look identical on a card can take wildly different amounts of time, and nobody knows which is which until it's done. So the chart, if anything, is optimistic, and a reader who takes it as the whole account will think getting below ninety percent busy is enough. A note about an approximation, not a complaint about anybody.

Miles: And the wider evidence for the practices.

Nora: Thinner than the book's confidence suggests. The best map of it is a pair of systematic reviews by Faustino and colleagues, Software Practice and Experience, twenty twenty two. The second goes hunting for implementation evidence behind the claimed benefits and finds thirty six documents containing sixty nine case studies. No trials. And the reviewers note in their own limitations that the case study authors frequently supplied no quantitative evidence for the benefits they reported.

Miles: So it's case reports again.

Nora: Case reports again, with the same publication problem underneath. Nobody writes up the transformation that went nowhere.

Miles: And the big survey everybody quotes.

Nora: Is a survey, and it's three episodes from here, so only what's needed. The research programme this book's community built reported, in its twenty nineteen edition, six years of data from over thirty one thousand professionals, about a thousand of them that year. Every figure is self reported by people who chose to fill in a questionnaire. A sample description rather than a criticism, and we'll do it properly when we get there.

Miles: So the tier statement.

Nora: Here it is. This is fiction written by consultants. Practitioner literature rather than peer reviewed research, reporting no study and claiming none. It dramatizes practices a community had already adopted, and what holds up underneath is borrowed: a queueing result from nineteen sixty one, and a constraint argument already dramatized in a novel three decades earlier. What's original here is the translation from a plant to a department, and translation is not evidence. It's very good teaching about ideas that were proved somewhere else.

Miles: Our half, and you've been sitting on it since the cold open.

Nora: Because the freeze is the thing. A work in progress limit is an executive function aid that nobody has to ask for, and I doubt the authors knew they'd written one.

Miles: Take it slowly.

Nora: Start with what the measures actually measure. Willcutt and colleagues, two thousand five, Biological Psychiatry. Eighty three studies pooled, executive function tasks, three thousand seven hundred and thirty four people with ADHD against nearly three thousand without. Impairment on every task, effect sizes in the medium band, point four six to point six nine, strongest on inhibition, vigilance, working memory and planning.

Miles: And the qualifier.

Nora: From the authors themselves. They conclude those weaknesses are neither necessary nor sufficient, meaning plenty of people with the diagnosis don't show them. A group difference on laboratory tasks. Not a test, and not a portrait of anybody.

Miles: And the other population.

Nora: Demetriou and colleagues, twenty eighteen, Molecular Psychiatry. Two hundred and thirty five studies, fourteen thousand and eighty one participants, autistic groups against non autistic ones on the same family of measures. Overall effect of point four eight, interval point four three to point five three, broad rather than concentrated in one sub domain. And the same caution by another route: they checked the measures for diagnostic sensitivity and only a few had it.

Miles: Two large pools, two populations, roughly the same medium difference.

Nora: And in both, part of what's measured is holding several intentions open at once and returning to the one you left. Which is the exact quantity a work in progress limit reduces. Not by asking the person to hold fewer. By making it against the rules for the board to hand them more.

Miles: And what does it cost when somebody has to come back to one?

Nora: Here's the finding for any manager who measures that by the clock. Mark, Gudith and Klocke, two thousand eight, at the human computer interaction conference. Forty eight participants, mostly German university students, an artificial email task, three conditions: uninterrupted, interrupted on topic, interrupted off topic.

Miles: And the interrupted groups were slower.

Nora: The interrupted groups were faster. Twenty point three and twenty point six minutes against twenty two point eight. And they reported more stress, more frustration, more time pressure and more effort. Small, artificial, students, so I'd stretch it no further than this: speed is a bad instrument for the cost of interruption, because people pay it where the stopwatch can't see.

Miles: Which brings you to the part about asking.

Nora: To the part about asking, and that's why the operating rule beats the accommodation. Lindsay and colleagues, twenty twenty one, Disability and Rehabilitation. A systematic review of disclosure and workplace accommodation for autistic people: twenty six studies, seven thousand and six participants, aged thirteen to seventy five, seven countries. Rates of disclosing, and of actually receiving adjustments, varied enormously. Reported benefits included acceptance and getting the adjustment. Reported limitations were stigma and discrimination.

Miles: And something with a number on the cost.

Nora: Von Schrader, Malzer and Bruyère, twenty fourteen, Employee Responsibilities and Rights Journal. An online survey, seven hundred and eighty responses, five hundred and ninety nine from people with a disability. Self selected and unusual: two thirds female, four fifths white, two thirds holding a degree. So generalize carefully. Inside that sample, seventy three percent named fear of being fired or not hired as a barrier, and nearly a quarter of those who disclosed reported long term negative consequences.

Miles: And a work in progress limit requires none of that.

Nora: None of it. It's a rule about the board. Nobody fills in a form, nobody explains to a manager how their head works, nobody weighs that worry against a thing they need. The person who needed the limit gets it, so does the person who didn't, and neither one is identified.

Miles: Say the caution, because I can feel one coming.

Nora: Two. First, nobody has run this study. There's no trial of work in progress limits with neurodivergent workers as an outcome, and what I've done is set two literatures side by side and draw a line between them. That's an argument. I'd rather label it than dress it up as a finding.

Miles: And number two.

Nora: The fence cuts both ways. Turning one person's narrow brilliance into a protected scarce resource is also how you build somebody who can't take a holiday, can't be promoted and can't be ill. The novel knows it, which is why the remedy isn't only the fence but the people writing down what he does. If a spike is what somebody's valued for, getting it out of their head is what stops it becoming a cage.

Miles: Fair enough.

Miles: One minute, in our frame. The Phoenix Project says a technology department is a factory, it has a slowest station, and most of its pain is unplanned work that earlier carelessness paid for. If you lead people who think differently, or you are one, here's Monday. Cap the number of things allowed to be open at once, and cap it on the board rather than in anybody's head, because that's an adjustment reaching everybody who needs it and costing nobody a disclosure. Name the four kinds of work out loud, because a department that can't say what it's doing will always call the loudest thing the most important. Protect the person the queue forms in front of, and in the same breath get what's in their head out of it, so protection doesn't turn into confinement. And remember ninety percent busy: the same people, working the same way, nine times slower. Slack isn't a reward for finishing. It's the mechanism.

Miles: Now put the machines against it.

Nora: Business first. The permanent warnings about this whole category sit at the very end now, past the music, in words that don't change, so I'll point at them and spend the time here on what this book gives us.

Miles: Go on.

Nora: Sort the tools into the four types of work and something useful falls out. They're strong on changes and internal projects, the tedious middle of both. They do nothing about unplanned work. And since unplanned work is generated by changes made carelessly, a tool that raises the rate of change without raising the rate of recording it manufactures the fourth type.

Miles: And the measurement question.

Nora: Is where it gets interesting, because there are two randomized trials now and they point in opposite directions. The first, twenty twenty three, recruited ninety five programmers through a freelancing platform, gave them one task, building a small web server, and randomly handed about half of them an assistant. That half finished fifty five point eight percent faster, interval from twenty one percent to eighty nine, and the biggest gains went to the least experienced.

Miles: And the trial going the other way.

Nora: Twenty twenty five. Sixteen experienced developers on open source projects they'd averaged five years on. Two hundred and forty six real tasks, each randomly assigned to permit the tools or forbid them. Completion time went up nineteen percent.

Miles: Up.

Nora: Up. And the line that belongs in this episode is what those same developers said afterwards. Asked to estimate, they reported they'd been about twenty percent faster. The clock said nineteen percent slower. They were not lying and they were not careless. They're skilled people whose sincere impression of their own speed sat roughly forty points away from the measurement.

Miles: And the two aren't actually a contradiction.

Nora: They describe two settings. A stranger's small task, where there's no context to lose, against your own codebase after five years, where the context is the asset. Between them they say the effect depends heavily on where you're standing, and anybody quoting one of those numbers at you has picked a side.

Miles: And what this book makes of that.

Nora: Everything. Erik's first demand is that Bill stop describing his department and go count it, because the felt sense of where the time went was wrong in every direction. A tool that makes work feel faster while the record says otherwise is exactly what he's warning about, and the remedy is unglamorous and unchanged. Measure it in your own building.

Miles: Leave us with something.

Nora: The book's own. Unplanned work is what makes planned work catch fire. These tools will let you start more of everything before lunch. Put the cap on first, because the one thing nobody can generate faster is the attention that finishes things.

Miles: That's The Phoenix Project.

Miles: Next time: The Lean Startup, by Eric Ries. Two days running we've had books about getting work through the building faster, and neither stops to ask whether anybody wanted the thing at the far end. Goldratt and Kim both hand you a plant that hums. Both take it as given that you already know what it ought to be making.

Nora: And Ries makes that the only question worth asking. That's a real argument, and it's about time somebody started it.

Miles: Nicely done, Nora. Thank you.

Nora: I'll have read it by then.

Miles: Go run something.

Miles: Stoar Consulting and Coaching makes Sixty Books. Logan Williamson is a Licensed Professional Counselor in Texas and Louisiana. What you just heard was a lesson about a book, for people who lead teams and people who belong to them. It was not coaching, it was not counseling, and no part of it was a reading of you.

Nora: Sources for everything are posted with the transcript on the episode page. Thirty three lands tomorrow.

Miles: Two disclosures. These voices were generated. Nobody is called Miles, nobody is called Nora, neither of them is licensed for anything, and nothing either of them said was about the person listening. AI helps draft every script here, and Logan Williamson, a Licensed Professional Counselor in Texas and Louisiana, at Stoar Consulting and Coaching, edits and approves what you get.

Nora: And one note about the AI segment. The evidence that these tools help neurodivergent people at work is early. There are controlled trials now, but they measure things like conversation skill rather than how a job actually goes, and for work itself most of what exists is individual providers reporting success. So take it as promising and not proven. The tools also get rebuilt every few months, so whatever you set up today is built on something that moves.