An Oxford tutorial for newsroom leaders who must learn faster than the news

Editor’s Note: Newsroom leadership rarely gets the deliberate practice it demands, and ComplexDiscovery OÜ built this tutorial to give it some. The second installment of The ComplexDiscovery Tutorial, following a first edition on cybersecurity, governance and eDiscovery, runs 21 contestable propositions about how newsrooms are led across three terms and reproduces three of them in full, essay and interrogation. The three anchor claims are that a newsroom’s real product is how fast it learns, that it owes its people safety before it owes the public a story, and that product thinking and AI now belong at the center of editorial judgment.

For editors, desk heads and media managers, the payoff is concrete. Each of them decides daily how to run a post-mortem, whether to push a tired reporter, and how far to let a metric or a model shape a call. The tutorial does not hand down answers; it shows how to test a leadership claim until its defensible core separates from its comfortable habit.

What to watch is whether newsrooms treat leadership as a craft to be learned rather than a rank to be reached. The propositions here are meant to be argued either way, and the most useful thing a reader can do is take the side they least believe and see whether it holds.

Industry News – Leadership Beat

An Oxford tutorial for newsroom leaders who must learn faster than the news

ComplexDiscovery Staff

An explanatory feature on the second installment of The ComplexDiscovery Tutorial, a leadership training exercise for editors and newsroom managers rather than a reported news feature. It describes the method, summarizes the source base and reproduces three illustrative worked examples, so it runs long by design and is both an article and a reproduced tutorial package. As first-party coverage, it describes a ComplexDiscovery OÜ publication rather than an outside development.

Picture a small room, a single essay read aloud, and one listener whose job is to find the sentence that does not hold. That is the Oxford tutorial, a form of teaching built to test ideas rather than transmit them, and it turns on a simple wager: a claim you can defend against someone trying to break it is worth more to you than a claim no one has questioned. ComplexDiscovery OÜ has taken that form and turned it on a subject that rarely receives it, which is how newsrooms are led.

Newsroom leadership is one of the few crafts still learned mostly by apprenticeship and accident, even as the environment around it changes by the quarter. Editors are promoted for their reporting and then handed responsibility for people, budgets, product and now artificial intelligence, with little deliberate practice at any of it. The tutorial offers that practice. It takes three live arguments about how newsrooms should be led and runs each through the same discipline: write the strongest case, then sit in the chair and answer for it.

What it is

The product is The ComplexDiscovery Tutorial, the same format ComplexDiscovery OÜ first pointed at cybersecurity, information governance and eDiscovery. This installment turns it on newsroom leadership. It is built around 21 contestable propositions across three academic terms, seven tutorials to a term plus a collections week, with a fully worked example for each term, a model essay paired with the interrogation that follows, so readers can watch the method at work rather than take its value on faith. The propositions concern how newsrooms are led rather than how stories are reported: the culture that lets a newsroom learn, the duty a newsroom owes its people, and the place of product thinking and artificial intelligence in editorial judgment.

The tutorial, softened

The tutorial is less a lecture than a disciplined conversation. The document frames its rhythm the way the Oxford original runs, a weekly argument between a tutor and a student built around a single essay of about 2,000 words. The essay specification is strict: one thesis stated in the opening paragraph, the strongest version of the opposing case presented fairly before it is answered, and a conclusion the argument actually earned rather than the one the writer preferred at the start. Every proposition is written to be argued either way, since a claim you can only defend is not worth a tutorial. What follows reproduces one representative session per term rather than a full term of them, so the three worked examples are a sample of the method, not its complete calendar. The aim is not to win but to find where the argument breaks, and to see whether it can be repaired without pretending the break did not happen.

Where the form comes from

The tutorial is old, and its age is part of the argument for using it. Oxford’s collegiate system took shape in the 13th century, and by the 15th century the role of the tutor was documented as responsibility for the conduct and instruction of younger colleagues. For most of that history the teaching happened through direct discussion, a student presenting ideas and an experienced scholar pressing on them. The individual, discussion-based tutorial reached its distinctive modern form in the middle of the 19th century. By tradition it is credited to Benjamin Jowett, the classicist and Master of Balliol College, who became Vice-Chancellor of the university in 1882 and whose Socratic method of one-on-one questioning spread as a pattern for the whole institution.

What survived those centuries is not a syllabus but a shape: one person makes a claim in writing, another person tries in good faith to take it apart, and the learning happens in the space between. That shape is indifferent to subject. It was built for classics and philosophy, and it works on any claim that can be stated and contested. Newsroom leadership is full of such claims, most of them inherited and few of them recently tested, which is what makes it a fair target for the method. The practical import for a modern editor is plain: the method is old, portable and available now, and it can be turned this week on the leadership assumptions a newsroom has never made argue for themselves.

A method, then three terms

The three tutorials are organized by the Oxford academic terms, which carry no meaning for most readers and deserve more explanation than a gloss. Oxford divides its year into three terms of eight weeks each, and the names come from the Christian liturgical calendar, a residue of the university’s medieval origins. Michaelmas, the autumn term, is named for the Feast of St Michael and All Angels on Sept. 29. Hilary, the winter term, is named for the feast of St Hilary of Poitiers, a fourth-century bishop and theologian, which traditionally falls on Jan. 14. Trinity, the spring term, is named for Trinity Sunday, which arrives eight weeks after Easter, in May or June.

The terms are used here as containers, autumn for the culture question, winter for the duty question, spring for the product and AI question, and nothing turns on the liturgy. Each term runs seven weekly propositions and a collections week that closes it with a synthesis essay or an oral defense. At Oxford, collections are internal college examinations, not counted toward the degree and normally sat at the start of term to review the previous term’s work; the tutorial borrows the name for its term-closing synthesis, the point at which the seven arguments are weighed together rather than one at a time. The order is deliberate: a newsroom has to be able to learn before it can be trusted to lead people, and it has to lead people well before it can govern the tools reshaping the work.

Michaelmas: from legacy culture to learning culture

What a leader practices here: building a newsroom that learns on purpose, through real listening, blameless post-mortems and safe experimentation.

Michaelmas, the autumn term, is about the culture that lets a newsroom learn. Week one opens on the proposition that a newsroom’s core product is not its journalism but the speed at which it learns, training the reader to separate an organization’s output from its capacity to keep producing under changing conditions, against Alexandra Borchardt’s argument for the Reuters Institute that leadership in news is a learnable craft whose first discipline is to listen, set beside Amy Edmondson’s account of learning organizations. Week two holds that the listening a reporter is trained to do is the first skill a leader abandons at promotion, and should be reclaimed and turned inward, reading Borchardt on listening against Jill Geisler’s management writing on hearing staff individually. Week three argues that communication is a management beat rather than an afterthought, and that ambiguity from the top is a leading source of distrust and burnout, anchored by Geisler’s guidance for the National Press Club Journalism Institute on leading through uncertainty. Week four presses the claim that a post-mortem which assigns blame trains a newsroom to hide its failures, drawing on Edmondson’s finding that error reporting depends on safety and on Google’s Project Aristotle, which identified psychological safety as the most important dynamic in effective teams.

The back half turns from habits to structure. Week five contends that innovation is a culture and not a side project, and that it dies when it is quarantined in a lab, reading the leaked 2014 New York Times innovation report, Strengthening Our Newsroom: Digital First, against the Online News Association’s leadership resources. Week six makes the precondition explicit, arguing that psychological safety is what allows experimentation rather than a reward for succeeding at it, with Edmondson and Project Aristotle as the readings. Week seven closes the teaching weeks with the proposition that a newsroom that cannot lose an argument cannot learn from one, turning the tutorial’s own method back on the newsroom that runs it. The term ends with a collections essay defending the view that a newsroom’s durable advantage is how fast it turns experience into retained practice, and naming which weekly positions the writer has since abandoned. One pressure point runs through all seven weeks: if learning is the product, a newsroom that learns fast but publishes little has optimized a loop around an empty center, and the tutor presses exactly there.

Hilary: human-centered leadership in high-stress newsrooms

What a leader practices here: leading people under pressure, through duty of care, calm and consistent crisis communication, and fair, transparent systems.

Hilary, the winter term, moves from culture to the people who live inside it. Week one, reproduced in full later, argues that a newsroom owes its people psychological safety before it owes the public a story, setting Amy Edmondson’s research, which established that psychological safety predicts whether people report errors and through that reporting whether a team learns, beside Joe Ruiz’s essay for the JSK Journalism Fellowships at Stanford on the toll of high-stress coverage. Week two holds that duty of care is a leadership obligation rather than a wellness benefit, reading Ruiz against the work of the Dart Center for Journalism and Trauma, the field’s long-running center on the psychological impact of covering violence, whose mission carried to the independent Global Center for Journalism and Trauma after the Dart Center closed in 2025. Week three argues that trauma-aware debriefs belong in the workflow as routine rather than as a favor granted on request, again drawing on the Dart Center’s training model. Week four reframes crisis leadership as calm, consistent communication and individual check-ins rather than visible heroics, anchored by the National Press Club Journalism Institute’s guidance on leading teams through uncertainty and fear.

The term’s second half turns to power and fairness. Week five holds that a newsroom’s systems for hiring, promotion, pay and assignment either dismantle inequity or entrench it, and that none of them is neutral, reading the Reuters Institute’s Changing Newsrooms 2023 survey, which found leaders conceding weaker performance on diversity at senior levels, against the Online News Association’s leadership resources. Week six argues that salary and promotion transparency is a trust mechanism rather than an exposure to be managed, with Geisler’s management writing as the primary text. Week seven closes, drawing on Ruiz, on the claim that if courageous journalism requires courage merely to show up to work, leadership has failed twice. The term ends not with an essay but with a viva, an oral defense of the proposition that the governance of a newsroom’s people is, in the end, the governance of its journalism. The pressure point is the word owes, and the tutor tests whether the ordering survives an urgent story and an exhausted reporter.

Trinity: product, audience and AI

What a leader practices here: governing product thinking, audience data and AI so they serve editorial judgment rather than quietly replace it.

Trinity, the spring term, carries the culture and the duty into the place the field is actually moving. Week one, reproduced later, argues that product thinking and AI belong at the center of editorial judgment rather than at its margins, with the Online News Association’s product and leadership resources and the Reuters Institute’s Changing Newsrooms survey as readings. Week two holds that journalism first is a north star for integrating product rather than a license to ignore it, reading Borchardt’s leadership guidance against the product literature. Week three argues that audience intelligence means understanding who a newsroom is failing to reach rather than counting who it already has, setting the Reuters Institute’s audience guidance beside Jay Rosen’s 2006 essay, The People Formerly Known as the Audience. Week four presses the claim that a vanity metric is an editorial decision no one will admit to making, a lesson in reading a number for the judgment buried inside it.

The closing weeks raise the stakes to the tools themselves. Week five argues that AI is infrastructure and that infrastructure is never neutral, reading Charlie Beckett’s JournalismAI report for Polis at the London School of Economics, New powers, new responsibilities, against the Changing Newsrooms findings on caution. Week six, contestable and hedged, holds that human oversight of AI can function as a liability shield rather than a safety control, testing whether the human in the loop can actually override the system or is only a signature at the end of it. Week seven projects the term onto a problem with no settled answer, arguing that the next trust crisis is not the convincing fake but the unlabeled authentic, reading the work of the Coalition for Content Provenance and Authenticity on content credentials against its own documented limits. The term ends with a synthesis essay tracing one argument across all three terms, from learning culture, through duty of care, to governed tools, and asking what it tells a working editor to do differently on Monday. The pressure point is the word center, which can be read as governance or as surrender, and the tutor does not let the writer keep the flattering reading for free.

The form in motion

Three worked examples follow, one per term. They are illustrative demonstrations, model essays paired with constructed interrogations, and the section that reproduces them opens with a full note on what is simulated and what is verified.

The first example defends the learning-culture proposition, then concedes under questioning that the learning rate is a thing to manage rather than a literal replacement for the story, and holds that weaker line on purpose. The second defends the duty-of-care proposition, distinguishes a demanding newsroom from a fearful one, and bounds the duty by what a leader actually controls. The third defends the product-and-AI proposition, keeps a fixed line of human accountability at the center, and argues that forces held at the margin are not governed but merely unmanaged.

Running it against your own work

The tutorial is built to be turned on the reader’s own decisions. Take a leadership claim you rely on, that the story comes first, that a metric is a distraction, that a tool is safe because a human signs off, and write the strongest case for it. Then put the sentence in front of a hostile examiner and ask what a careful critic would say next. The payoff is not a verdict but a habit: the claims that survive that treatment are defensible, and the ones that only feel familiar are exposed before an audience or a staff exposes them.

Return to that small room and the single listener. The point of the tutorial was never the verdict the student reached; it was the chair the student sat in, the one seat where a comfortable claim has to answer for itself out loud. Newsroom leadership offers few such seats, which is why so many of its claims go untested until a resignation or a correction tests them instead. Put your own defaults in the chair while the stakes are only an argument. The ones still standing at the end of the hour are the ones worth leading with.

The tutorials in full

The following worked examples are illustrative demonstrations provided by ComplexDiscovery OÜ. Each pairs a model essay with a constructed interrogation: a written simulation of a tutorial exchange, voiced by composite tutor and student figures created to demonstrate the method. They are not transcripts of a live session between named individuals. The authors, studies and organizations cited within the essays are real and were verified before publication; the propositions are written to be argued either way and are not the editorial positions of ComplexDiscovery OÜ.

Tutorial one, Michaelmas: “A newsroom’s core product is not its journalism but the speed at which it learns”

Explainer

This example trains the separation between an organization’s output and its capacity to keep producing that output under changing conditions. Watch two things: whether the writer can hold the claim without dismissing journalism itself, and whether learning is defined tightly enough to be managed rather than merely admired.

The essay

The following essay defends the proposition. The writer’s task is to build the strongest case, then defend it under questioning.

A newsroom is judged by what it publishes, and it should be. But the thing that determines what it can publish next year is not this year’s best story. It is the set of habits that let the newsroom notice what it got wrong, adjust and try again before its competitors or its critics force the issue. On that view, the core product of a newsroom is not journalism as an artifact but the rate at which the organization turns experience into improved practice.

The case rests on the environment. Audiences, platforms and tools now shift faster than any single editorial strategy can be perfected. Alexandra Borchardt, writing for the Reuters Institute, argues that leadership in news is a craft that can be learned and taught, and that its first discipline is to listen first and talk less, to the team, to the audience and to the outside world. A newsroom that listens on a schedule, and acts on what it hears, is running a learning loop whether or not it calls it one.

The practical content is unglamorous. It is the regular one-on-one that surfaces a problem before it becomes a resignation. It is the post-mortem that asks what the process allowed rather than who to blame. The Reuters Institute’s account of newsroom leadership from the 2023 International Journalism Festival stresses preparation before a crisis rather than improvisation during one, which is another way of saying the learning has to happen ahead of the pressure. A newsroom that debriefs a failed launch and changes one workflow has learned; a newsroom that praises the launch and moves on has not.

Consider what the claim looks like inside a working week. A desk tries a new live-blog format for an election night. In a newsroom organized around stories, the format is judged by whether that night went well, and the question ends there. In a newsroom organized around learning, the same night generates a second artifact, a short written account of what the format assumed about the audience, what actually happened and what the desk would change. That account is worth nearly as much as the coverage, because the coverage is spent and the account compounds. The leaked 2014 New York Times innovation report, Strengthening Our Newsroom: Digital First, described the opposite condition, a great newsroom whose digital learning was treated as a bolt-on and whose hardest-won lessons never became shared practice. The report’s force was that it named, from the inside, an outlet publishing brilliantly and learning slowly.

What would it mean to manage the learning rate directly rather than admire it after the fact? It means treating each substantial decision as a small experiment with a stated expectation, then checking the result against the expectation on a fixed schedule. A newsroom that starts a newsletter can guess at its audience, or it can write down who it expects to reach and why, ship it, and read the gap between the guess and the result as information rather than as a verdict on anyone’s talent. The difference is not effort. It is whether the newsroom has built a place to put what it learns, so the next decision begins from the last one rather than from scratch. Hero culture keeps that knowledge in the heads of a few stars, which is why it walks out the door when they do.

The strongest case against the proposition is worth stating in full, because it is mostly right. A newsroom is not a laboratory, and its readers do not subscribe to its learning habits. They subscribe to its journalism. An outlet that produced the finest investigative work of its generation while learning nothing about its own processes would still have discharged its obligation to the public, and an outlet that learned feverishly while publishing nothing would have failed that obligation completely. Output is the thing with public value; the learning rate has value only because it feeds the output. To call the learning rate the core product is, on its face, to confuse the engine with the cargo.

The answer is not to deny the objection but to locate what it misses. The objection is decisive against the literal claim and powerless against the useful one. No one proposes to stop publishing in order to learn better. The proposition is a claim about which quantity a leader should manage, given that the output is already the thing everyone watches. A newsroom’s published work is a lagging measure, visible only after the decisions that produced it have been made. By the time a declining franchise shows up in the traffic, the habits that let it decline have been in place for years. Managing the learning rate is managing the leading indicator, the pace at which the newsroom updates the practice that will produce next year’s work. The cargo is what matters; the engine is what a leader can still do something about today.

There is a second reason the learning rate deserves the leader’s attention, which is that it is fragile in a way the output is not. A good story, once published, exists. A newsroom’s capacity to produce the next one lives in habits, relationships and shared understanding that can be lost without anyone deciding to lose them. Google’s Project Aristotle, examining its own teams, found psychological safety to be the most important dynamic in effective teams, which is another way of saying that the conditions for learning are interpersonal and therefore perishable. A reorganization, a few departures, a season of fear after layoffs, and the newsroom still publishes while quietly losing the ability to improve. Naming the learning rate as the product puts a number on something that otherwise erodes in silence.

The practices that raise the learning rate are unglamorous and specific, which is why they are so often skipped. Borchardt’s guidance for the Reuters Institute puts listening first, and a leader who listens on a schedule, in standing one-on-ones and in structured sessions with the newsroom and the audience, is running the intake stage of a learning system. The output stage is equally concrete: a decision memo that states why a change is being made, so that when the result arrives there is a recorded expectation to compare it against. Jill Geisler’s management writing makes the same point from the other side, that ambiguity from the top is not a neutral absence but an active source of distrust and burnout. A newsroom that cannot say why it is doing something cannot later say what it learned from doing it, because it never recorded what it expected.

It is worth being precise about the difference between a learning newsroom and a merely busy one, because the two are easy to confuse and the confusion is expensive. Agility, velocity and a full calendar of experiments are not learning; they are its raw material at best and its counterfeit at worst. Learning is the closing of the loop, the moment a result is compared against a stated expectation and a practice changes as a consequence. A newsroom can ship a redesign every quarter and learn nothing, if no one wrote down what each redesign was supposed to achieve. The discipline is not doing more things. It is doing a smaller number of things with a stated hypothesis and an honest reckoning afterward, which is slower to start and far faster to compound.

The proposition also has to survive its own success, because a learning rate that becomes a target stops measuring learning. Once a newsroom rewards people for running experiments, it will get experiments, worthwhile or not, and the debrief becomes a form to complete rather than a question to answer. The safeguard is to keep the judgment of what is worth learning in editorial hands and to resist turning the learning rate into a dashboard figure a manager can optimize. The proposition is a claim about where a leader should look, not a number to be gamed, and a newsroom that forgets the distinction will have proved the objection it was meant to answer.

The proposition invites a counterfeit worth flagging before it spreads. A newsroom can perform learning, running experiments because experimentation is fashionable and debriefing everything until the debrief is a ritual no one reads, and mistake the motion for the thing. The learning rate is not a volume of activity. It is the rate of useful change, and useful is the hard word, because it can be judged only against the mission the newsroom exists to serve. That judgment is editorial, and it is the one part of the loop that cannot be handed to a process.

The proposition is deliberately overstated, and its weakness is plain. A newsroom that learns quickly but publishes nothing worth reading has optimized a loop around an empty center. Learning is a means. Yet the overstatement earns its place, because most newsrooms already know how to value the story and do not yet know how to value the system that produces it. Naming the learning rate as the product forces a question hero culture never asks: not whether this was a good story, but whether the newsroom got better at making good ones.

The interrogation

Tutor: You say the core product is the learning rate. If I shut down your newsroom’s output for a year to perfect its learning habits, have I improved the product?

Student: No. I have destroyed it. The learning rate only counts as a product because it is measured against output. I would restate the claim: the learning rate is what should be managed, because the output is already what everyone attends to.

Tutor: That is a retreat. Managed is weaker than the core product.

Student: It is a retreat, and I will hold the weaker line. The strong version was a corrective, not a literal claim about value. What I will not concede is that output alone is a sufficient object of management, because output is a lagging measure.

Tutor: Borchardt says leadership is a learnable craft. That is a claim about individuals. You are using it to support a claim about systems. Is that a slide?

Student: It is a fair challenge. I am extending her from the leader to the organization. The extension holds only if individual learning is captured somewhere the next person can reach it, in a documented decision or a changed workflow. If it lives only in one editor’s head, a person has learned and the newsroom has not.

Tutor: Then your real proposition is about institutional memory, not speed.

Student: Speed and memory are linked. A newsroom can learn something and lose it, or learn it and keep it. I would rather defend the version that says a newsroom’s advantage is how fast it turns experience into retained practice.

Tutor: Retained practice could describe a bureaucracy that has learned to protect itself. How do you tell learning from ossification?

Student: By whether the retained practice is still tested. A learning system keeps a habit because it keeps earning its place; a bureaucracy keeps it because removing it is inconvenient. The tell is whether anyone is allowed to run the experiment that would kill the rule.

Tutor: Give me the failure mode of your own proposal. A newsroom that worships the learning loop, what does it get wrong?

Student: It mistakes motion for learning. It runs experiments because running experiments is the culture, and it debriefs everything, including things that did not matter, until the debrief is a ritual no one reads. The correction is to be as disciplined about which experiments are worth running as about running them. A learning rate is not a volume of activity; it is the rate of useful change, and useful is the hard word.

Tutor: Who decides useful?

Student: The editor, against the mission, which is the one judgment I do not want to automate. The loop can tell you whether a change worked. It cannot tell you whether the change was worth wanting. That is why the proposition is a claim about capacity and not a proposal to replace editorial judgment with a dashboard.

Tutor: Last question. If psychological safety is what lets people report failure honestly, is your proposition really about learning, or about culture wearing a learning costume?

Student: They are the same claim seen from two sides. There is no learning rate without honest reporting of failure, and no honest reporting without safety. I accept that culture is the mechanism and learning is the outcome, and I defend the proposition on the outcome.

Tutorial two, Hilary: “A newsroom owes its people psychological safety before it owes the public a story”

Explainer

This example trains the ordering of obligations, duty of care against mission. Watch whether the writer treats safety as an alternative to rigor or as a precondition for it, and whether the word owes survives a case where the story is genuinely urgent.

The essay

The following essay defends the proposition. The writer’s task is to build the strongest case, then defend it under questioning.

The phrase the story comes first is a badge of honor in newsrooms, and it has justified a good deal of harm. The proposition here inverts the priority: before a newsroom is owed anything by its staff, it owes them a workplace where reporting the truth, including the truth about the newsroom, does not require courage. The claim is not that the story does not matter. It is that a newsroom which extracts the story by spending its people is borrowing against a fund it cannot replenish.

The strongest support is empirical. Amy Edmondson’s psychological-safety research, including her 1999 Administrative Science Quarterly study of 51 work teams in a manufacturing company and her earlier work on medical-error reporting, established that psychological safety, a shared belief that a team is safe for interpersonal risk-taking, predicts whether people report errors and, through that reporting, whether the team learns. A newsroom is an error-generating environment under deadline. Without safety, the errors are hidden rather than fixed, and the hiding is itself a threat to the public the newsroom serves.

The duty has a human floor as well as an instrumental one. Writing for the JSK Journalism Fellowships at Stanford, Joe Ruiz argues that newsrooms are losing good people because the support offered to those covering conflict, disaster and political volatility has not matched the toll. Jill Geisler, in guidance for the National Press Club Journalism Institute, presses leaders to listen to staff individually rather than assume a single reaction to a hard moment, and to help people reconnect with why the work matters. Both describe the same obligation: the newsroom acts first, and the demand on the reporter comes second.

The obligation has a concrete shape, which is what separates it from a slogan. It looks like a trauma-aware debrief after intense coverage, offered as routine rather than as a favor granted to whoever thinks to ask. It looks like an escalation path a reporter can use without wondering whether using it marks them as fragile. It looks like a promotion process whose criteria are written down, because a newsroom that cannot say how it decides who rises is asking its people to trust that the decision is fair while giving them no way to check. Each of these is a system, and systems are what a leader can actually change. The alternative is to ask individuals to be resilient in place of a structure that should have carried them, and to call the result toughness.

The strongest objection is not indifference but a rival ideal, and it deserves a fair hearing. On this view, journalism is a calling with a cost, and the people who choose it accept that the work sometimes comes before their comfort. War correspondents, investigative reporters and breaking-news desks do difficult things because the public needs them done, and a newsroom that put staff wellbeing first would, on this account, produce softer, slower, less consequential work. Pressure forges quality. A newsroom that will not ask hard things of its people has confused kindness with the abdication of its public purpose.

The objection is right that the work is hard and wrong that hardness and fear are the same thing. Edmondson’s research is explicit that psychological safety is not the lowering of standards; it is the condition under which high standards can be met without concealment. A demanding assignment handed to a reporter by an editor they trust, with support around it and honesty about the toll, is not a failure of duty of care. It is duty of care operating at full strength. What the proposition refuses is not difficulty but the arrangement in which difficulty is extracted through fear, and in which the reporter who says the assignment is breaking them learns that saying so marks them as weak. The work of the Dart Center for Journalism and Trauma, which spent decades studying the psychological impact of covering violence, rests on that distinction: reporters can do the hardest work in journalism when the organization treats the harm as real and builds for it, and the same work destroys people when the organization treats the harm as a private weakness.

There is an instrumental case and a human case, and the proposition needs only the first to stand, though it rests on both. The instrumental case is that a newsroom runs on people volunteering information the organization needs and no one can compel, a doubt about a source, an error in a chart, a story that should be killed, and that fear silences exactly that flow. The human case is simpler and older, that an organization which spends its people to make its product owes them something in return, and that the something is a floor rather than a perk. A leader who accepts only the instrumental case will still arrive at most of the same practices, because the arrangements that make people safe enough to speak are the same ones that make a workplace humane.

The most common way the duty is breached is not cruelty but arithmetic. A newsroom sets its staffing on the assumption that people will absorb the overflow, that a team built for calm weeks will find a way through the hard ones, and that the finding-a-way is a private matter for each reporter. Dressed as an emergency, the resulting crunch looks like the rare night the proposition concedes must sometimes come first. It is not. It is a standing condition disguised as a series of exceptions, and the disguise is what lets a newsroom believe it honors a duty it has quietly defunded. A leader who owes safety first has to look at the staffing plan and ask whether the emergencies it produces are truly unforeseeable or merely unfunded.

The duty extends past the newsroom’s stress into its fairness, because a workplace people do not trust is not one they feel safe in. Systems for promotion, pay and assignment either earn that trust or corrode it, and there is no neutral setting. The Reuters Institute’s Changing Newsrooms survey found leaders conceding that their diversity efforts weakened at senior levels, which is where the consequential decisions are made. A promotion process whose criteria are unwritten asks people to trust that it is fair while giving them no way to verify the claim, and an assignment desk that hands out the desirable and the punishing work by habit will reproduce whatever inequities the habit contains. Salary and promotion transparency is uncomfortable precisely because it converts a private assurance into a checkable one, which is the same reason it builds the safety the proposition demands.

It helps to say what the proposition is not, because its critics tend to attack a softer version. It is not a claim that reporters should be shielded from difficulty, protected from criticism or handed a veto over hard assignments. A newsroom that confused safety with comfort would fail its public and, in the end, its people, who did not enter journalism to be coddled. The claim is narrower and harder: that the difficulty should be real difficulty, chosen and supported, rather than fear manufactured by a workplace that treats its people as an inexhaustible input. The demanding editor and the safe newsroom are not in tension. The editor who is exacting and trusted is how safety and high standards arrive together.

Picture the moment the proposition is really about. It is late, a story is nearly ready, and the reporter who has carried it for a month is visibly past empty. The newsroom that owes safety first does not necessarily pull the story, because sometimes the story cannot wait. What it does is make that night a rare exception rather than a normal Tuesday, ensure the exhaustion was not manufactured by a staffing plan that assumed heroics, and see that recovery follows. The newsroom that owes the story first treats the same night as ordinary, repeats it until the reporter leaves, and calls the departure a lack of resilience. Same night, two newsrooms, and the difference is entirely in which obligation was treated as the default.

The proposition overreaches, and the overreach is where it earns its keep. Taken literally, before would license a reporter to refuse an urgent assignment on the ground that the newsroom had not first perfected its duty of care, which no functioning newsroom could accept. The honest version concedes that safety and story are not always sequential; some stories cannot wait. What the proposition refuses is the standing arrangement in which the story is always first and the duty of care is whatever is left over. Reverse the default, and the hard cases become exceptions to argue rather than a culture to survive.

The interrogation

Tutor: Before is doing a lot of work. A source is about to go on the record and your reporter is exhausted. Do you owe the reporter rest before you owe the public the story?

Student: Not in that instant. In that instant the story may have to come first. What the newsroom owes the reporter is that the instant is rare, that it was not manufactured by chronic understaffing, and that there is recovery after it. The before is about the default, not the emergency.

Tutor: Then you have conceded the proposition. The story comes first when it matters most.

Student: I have conceded the literal reading and kept the operative one. The proposition is a claim about which obligation is the baseline and which is the exception. Newsrooms have run the exception as the baseline for a century. Naming safety as the thing owed first is how you stop doing that.

Tutor: Edmondson’s early work looked at hospital error reporting and her 1999 study at manufacturing teams. Are newsrooms enough like those settings for the finding to carry?

Student: The mechanism transfers better than most. Her claim is that people withhold information about error when they fear the interpersonal cost. Newsrooms run on people volunteering information: a doubt about a source, a mistake in a chart, a story that should be killed. If the cost of speaking is high, the information does not arrive, and the public inherits the error.

Tutor: You lean on Ruiz and Geisler. They advocate for this position. Where is the voice that says a hard newsroom produces hard journalism?

Student: It deserves an answer rather than a dismissal. The claim is that pressure forges quality, and sometimes it does. My response is that pressure and fear are not the same thing. A demanding standard applied by someone you trust is safety, not its opposite. Edmondson is explicit that psychological safety is not lowered standards; it is the condition under which high standards can be met without concealment.

Tutor: So a safe newsroom could still be a demanding one.

Student: It has to be. A safe and undemanding newsroom is pleasant and produces little. The proposition is not comfort first. It is safety as the floor that lets you raise the ceiling.

Tutor: You keep saying safety is not comfort. Draw the line. A reporter says a hard assignment is bad for their wellbeing. Do you owe them a different assignment?

Student: You owe them a real conversation and an honest answer, not automatic relief. Safety is the freedom to raise the concern without penalty and to be heard. It is not a veto over the work. Sometimes the answer is that the assignment stands and here is the support around it; sometimes the concern reveals something the desk missed. The duty is that the concern can be spoken and is weighed, not that the outcome is decided in the reporter’s favor in advance.

Tutor: That sounds like it could justify almost any decision, so long as you held the conversation.

Student: It could be abused that way, which is why the test is not whether a conversation happened but whether concerns ever change outcomes. A newsroom where raising a concern never alters anything has procedure without safety. If people learn that speaking up is theater, they stop speaking, and you are back to the hidden error the proposition was built to prevent.

Tutor: Final challenge. If you owe safety first, and a leader cannot provide it because the business is failing, has the leader failed morally or only practically?

Student: Practically, and the distinction matters. Duty of care is bounded by what the leader controls. The obligation is to act on the controllable, honest workloads, fair process, individualized listening, and to be candid about the rest. A leader who does the controllable and names the constraint has met the duty. One who hides behind the constraint to avoid the controllable has not.

Tutorial three, Trinity: “Product thinking and AI belong at the center of editorial judgment, not at its margins”

Explainer

This example trains the integration of product, audience data and AI with editorial judgment without collapsing one into the other. Watch whether center is read as governance or as capitulation, and whether the writer keeps a clear line of accountability that never moves onto the tool.

The essay

The following essay defends the proposition. The writer’s task is to build the strongest case, then defend it under questioning.

Newsrooms have long treated product, audience data and now AI as adjacent departments: useful, occasionally threatening, and safely downstream of the real work of editing. The proposition argues that this arrangement is no longer tenable, and that product thinking and AI have moved to the center of editorial judgment. Journalism first remains the north star. The claim is that serving that mission now requires the disciplines that were treated as peripheral.

Product thinking, at its core, is a demand for clarity: what problem, for which user, measured how. Applied to editorial work, it sharpens rather than dilutes. A section that cannot say who it is for, or how it would know it succeeded, is not protecting editorial independence; it is declining to be accountable. The Online News Association’s leadership and product resources treat this literacy as part of the modern newsroom’s basic equipment rather than a concession to the business side.

The strongest objection reverses the proposition’s own logic. Centering product, audience and AI, on this view, is precisely how newsrooms lose themselves. The metric that starts as a servant becomes the master; the tool adopted for efficiency begins to decide what gets covered; the product manager’s user, defined by engagement, quietly replaces the editor’s public, defined by need. Keep these forces at the margin, the objection runs, and editorial judgment stays sovereign. Bring them to the center and the newsroom has handed its purpose to the parts of it that were never supposed to be in charge.

This is the serious case, and the reply is not that it is wrong but that its remedy fails. Marginalizing these forces does not keep them from shaping the work; it only keeps anyone from governing how they shape it. A metric held at arm’s length still drives behavior, because people optimize for what is measured whether or not the measure is honored openly. A tool adopted quietly still changes which stories are cheap to tell. The drift the objection fears is real, and it happens most reliably in the newsrooms that refuse to look at it directly. Judgment cannot govern what it holds at a distance. The choice is not between purity at the margin and corruption at the center; it is between forces that shape the work unwatched and forces that shape it under a named person’s supervision.

Audience intelligence is the same argument in data form. The Reuters Institute’s leadership guidance urges editors to understand audience behaviors, needs and the gaps in who is not being reached, rather than chase volume. That is a shift from vanity metrics to editorial hypotheses tested against evidence. It does not hand the assignment desk to a dashboard; it asks the desk to be honest about whether its assumptions about the audience are true.

Audience intelligence is easy to caricature as the surveillance of readers in service of clicks, which is why the framing matters. Jay Rosen’s 2006 essay, The People Formerly Known as the Audience, named the shift that makes the caricature obsolete: the people once on the receiving end of a one-way broadcast are now participants who can talk back, organize and route around the newsroom entirely. Understanding them is not a marketing function bolted onto journalism; it is part of knowing whether the journalism is reaching anyone, and whom it is leaving out. The discipline the Reuters Institute describes, attending to who is not being reached rather than flattering totals, is an editorial question wearing a data coat. A newsroom that refuses to ask it is not protecting its independence. It is declining to find out whether its work lands.

AI is where the proposition is most contested, and the essay concedes the danger plainly. The Reuters Institute’s Changing Newsrooms 2023 survey found leaders both drawn to generative AI’s efficiencies and wary of its risks, which is the correct posture. The defensible position treats AI as infrastructure: support for fact-checking, summarization, translation and workflow, under transparent use policies, with a named human accountable for every error. On that model, AI at the center does not mean AI in charge. It means the leader sets the guardrails deliberately rather than letting the tools diffuse into the workflow ungoverned.

The guardrail is not a slogan about keeping humans in the loop; it is a specific arrangement. A transparent use policy states where the tools may be used and where they may not, published so an audience can see it. A named owner is accountable for each output, so that an error made with a tool is owned by a person and not excused by the tool. A standing review asks whether the tool is still earning its place, because a tool adopted for efficiency can quietly begin shaping which stories are easy to tell and therefore which get told. Human oversight can function as reassurance rather than as control; the difference is whether the human can actually override the system and is expected to, or is only a signature at the end of a process no one could realistically stop.

The claim that AI is infrastructure is doing specific work, and Charlie Beckett’s JournalismAI survey for Polis at the London School of Economics supports its shape rather than its comfort. That global study found AI already embedded in journalism and unevenly distributed, used mostly to extend the reach and efficiency of human journalists rather than to replace them, and paired the finding with a warning that new powers arrive with new editorial and ethical responsibilities. Infrastructure is the right word precisely because it is not neutral. A bridge decides where a road can go. A summarization tool decides which stories are cheap to produce and, over time, which get produced. Naming AI as infrastructure at the center is a way of insisting that a person owns the decisions the infrastructure is otherwise making on its own.

The final reason to govern these tools at the center rather than the margin is that the next crisis of trust will not announce itself as a machine. The Coalition for Content Provenance and Authenticity has built a technical standard for content credentials, cryptographically signed metadata that records where a piece of media came from and how it was edited. It is a serious response to synthetic media, and its own documentation is candid that it cannot certify a claim is true and that its protections can be stripped or forged. The lesson for a newsroom is that the hard problem is shifting from detecting the convincing fake to vouching for the authentic, and that vouching is an editorial act no tool performs on its own. A newsroom that has pushed provenance to the margin as a technical matter will meet that crisis without having decided who owns it.

The insistence on the center is finally a claim about accountability, which is where the abstract argument becomes concrete. Consider a fact-check that a tool returned with confidence and a desk accepted without a second source, published and wrong. In a newsroom that kept the tool at the margin, the failure diffuses: the tool was only a helper, no one quite owned its output, and the lesson evaporates. In a newsroom that placed the tool at the center under a named owner and a written use policy, the same error has an address. Someone is accountable, the policy is revised, and the newsroom has learned rather than merely apologized. Center, in the sense the proposition intends, is nothing more mystical than the condition under which a machine-assisted mistake is owned by a person.

Journalism first, properly understood, is what makes the integration safe rather than what forbids it. The phrase is often used as a wall, a way to keep product, audience and AI outside the room where the real decisions are made. On the proposition’s account it is a compass instead, the standard against which every product choice, every metric and every tool is judged. A metric that serves the mission is adopted; one that serves only engagement is refused. A tool that extends the reach of human reporting is welcomed; one that quietly substitutes for it is not. The Online News Association’s treatment of product literacy as basic newsroom equipment assumes exactly this, that a newsroom can hold the disciplines close without surrendering to them, provided it keeps the mission as the thing the disciplines exist to serve.

The proposition’s weakness is that center can be read as surrender, and some newsrooms will read it that way and let the metric or the model make the call. That misreading is real, and the essay does not pretend otherwise. The reply is that keeping these forces at the margin does not prevent the surrender; it only leaves it unmanaged. Judgment cannot govern what it holds at arm’s length. Bringing product, audience and AI to the center is the precondition for editorial judgment to stay in command of them.

The interrogation

Tutor: You want AI and product at the center but insist judgment stays in command. What stops center from meaning the metric decides?

Student: A line of accountability that never moves onto the tool. The rule is that a person owns every published decision and every error, including errors produced with AI assistance. If no one can be named, the tool is not at the center of judgment; it has replaced it, and the proposition is violated.

Tutor: That is a governance answer. But once a metric is central, people optimize for it. Do you really control what you have placed at the center?

Student: Not fully, which is why the choice of metric is itself an editorial act. If the central measure is reach at any cost, judgment has already been outsourced. The Reuters Institute’s point is to move to measures tied to audience need and to editorial hypotheses. A central metric that encodes the mission pulls toward the mission; one that encodes volume pulls away. The discipline is choosing the measure, not avoiding measurement.

Tutor: You cite the Online News Association and the Reuters Institute for product and audience, and the Changing Newsrooms survey for AI. Two of those advocate for the integration you defend. Is your evidence base tilted?

Student: It is, and I will name the tilt. These are practitioner and research bodies with a stake in newsrooms adopting these disciplines. The Changing Newsrooms finding is the useful check, because it reports leaders as cautious, not enthusiastic. My argument does not need enthusiasm. It needs the claim that these forces already shape the work, which even the cautious leaders concede. Given that they shape it either way, the only question is whether they are governed.

Tutor: On AI as infrastructure. Fact-checking support that is wrong in a confident voice is worse than none. Does infrastructure understate the risk?

Student: It does if it implies neutrality. Infrastructure is not neutral; a bridge decides where you can cross. That is why the model requires transparent use policies and a named human owner rather than trust in the tool. The proposition is not that AI is safe at the center. It is that AI is already load-bearing in many newsrooms and is safer governed at the center than diffused at the margins where no one signed off on it.

Tutor: Suppose a small newsroom cannot afford product staff or AI governance. Is the proposition a luxury?

Student: The tools scale down further than the discipline does. A two-person newsroom cannot hire a product manager, but it can still ask who a project is for and how it will know it worked, and it can still write one page on how it uses AI and who is accountable. The proposition is about where judgment sits, not about headcount. The cost of the discipline is attention, which every newsroom has to allocate anyway.

Tutor: You concede AI can shape which stories are easy to tell. If the tool biases the newsroom toward the summarizable and away from the slow, has the margin already won?

Student: It has if no one is watching for exactly that. The bias you name is real and quiet, which is the argument for the center rather than against it. A tool at the margin produces that drift and no one is assigned to notice. A tool at the center comes with the standing question of what the newsroom is now doing less of because the tool made something else cheap. Governance does not remove the bias. It makes the bias someone’s job to see.

Tutor: Name the thing you would refuse to do with AI, whatever the efficiency.

Student: Decide what deserves coverage, and pass off machine-generated work as human reporting. The first is the editorial act itself, so delegating it is not efficiency but abdication. The second breaks the trust the whole enterprise runs on. Everything between those, drafting help, translation, sifting documents, is negotiable and governable. Those two are the line, and a use policy that will not name its own lines is not a policy.

Tutor: Final question. If product, audience and AI are all central, what is left that is distinctly editorial?

Student: The decision about what is worth the public’s attention and why, which none of the three can make. Product tells you whether you served a user; audience tells you who you reached; AI tells you what it can produce. None tells you what matters. Bringing them to the center frees editorial judgment to do the one thing that is irreducibly its own, by handing it better instruments for everything else.

An invitation

The tutorial ends without a tidy verdict, and that is the point. None of the three reproduced propositions is settled here. The learning-rate claim retreats under pressure to a claim about retained practice; the duty-of-care claim concedes the emergency and keeps the default; the product-and-AI claim holds the center only by pinning accountability to a person. Each is stronger for having been argued against.

Treat this as an invitation to argue the side you do not believe. If you are certain the story comes first, write the case that a newsroom owes its people safety before it owes the public a story, and see where it forces you. If you are sure AI belongs at the margin, defend it at the center and find the line you will not cross. The propositions are built to be taken either way.

The questions are free. The thinking is the work. A leadership claim that can survive a hostile examiner is one you can act on; one that cannot is worth finding out about before it costs you a story or a colleague.

Sources and reference materials

References are cited in full below. Where a freely accessible public version exists, the title links to it.

ComplexDiscovery OÜ

• ComplexDiscovery OÜ, The ComplexDiscovery Tutorial (editorial series), complexdiscovery.com.
• ComplexDiscovery Staff, An Oxford tutorial for cybersecurity, governance and eDiscovery (the first installment of The ComplexDiscovery Tutorial), ComplexDiscovery OÜ.

The Oxford tutorial and its terms

• David Palfreyman, ed., The Oxford Tutorial: Thanks, You Taught Me How to Think, OxCHEPS Occasional Paper No. 1, Oxford Centre for Higher Education Policy Studies, 2002, available here.
• “Michaelmas Term,” Oxford Reference (Oxford University Press), on the term names and their origins in the liturgical calendar.

Research and institutional guidance

• Alexandra Borchardt, How to be an excellent leader in news, Reuters Institute for the Study of Journalism, RISJ Review.
• Reuters Institute for the Study of Journalism, Five tips on newsroom leadership from the International Journalism Festival 2023.
• Reuters Institute for the Study of Journalism, Changing Newsrooms 2023: Media leaders struggle to embrace diversity in full and remain cautious on AI disruption (survey of 135 senior news leaders across 40 countries), 2023.
• New York Times, Innovation (“Strengthening Our Newsroom: Digital First”), internal newsroom innovation report, 2014.
• Charlie Beckett, New Powers, New Responsibilities: A Global Survey of Journalism and Artificial Intelligence, JournalismAI, Polis, London School of Economics, 2019, available here.

Professional and fellowship commentary

• Joe Ruiz, What do we as newsroom leaders owe our people? JSK Journalism Fellowships at Stanford.
• Jill Geisler, Tips for newsroom leaders to help guide their teams through this moment, National Press Club Journalism Institute, Nov. 6, 2024.
• Jill Geisler, Good managers are guides, not guards, National Press Club Journalism Institute.
• Jay Rosen, The People Formerly Known as the Audience, PressThink, June 27, 2006.

Scholarship and research

• Amy C. Edmondson, The Fearless Organization: Creating Psychological Safety in the Workplace for Learning, Innovation, and Growth (Hoboken, N.J.: Wiley, 2018).
• Amy C. Edmondson, “Psychological Safety and Learning Behavior in Work Teams,” Administrative Science Quarterly 44, no. 2 (1999): 350-383.
• Google, Project Aristotle (re:Work, “Understand team effectiveness”), identifying psychological safety as the most important dynamic in effective teams, 2012 onward.

Standards and frameworks

Coalition for Content Provenance and Authenticity (C2PA), Content Credentials technical specification for content provenance, c2pa.org.

Assisted by GAI and LLM Technologies

Additional reading

• An Oxford tutorial for cybersecurity, governance and eDiscovery
• Estonia aims to be first to give AI agents official digital IDs
• Parliament hits pause on high-risk AI rules and bans nudifier apps
• Europe’s AI labeling rules arrive with a voluntary code and a hard deadline
• AI is becoming an advisor, and it is rewriting how buyers find you
• What marketing confidence actually looks like
• When credibility comes first, capability sells itself: The case for news-led marketing
• The one question that reveals whether your marketing plan is actually a plan
• How Prompt Marketing Is Redefining Thought Leadership In The AI Era
• Raising The Age Ceiling: How AI Is Extending Executive Leadership
• Staying Curious: One Practical Defense Against Creative Burnout
• From Longbows To AI: Lessons In Embracing Technology
• 20 Ways Creative Professionals Battle Burnout And Find Fresh Ideas
• 14 Points For Brands To Consider Before Making Sociopolitical Statements

Source: ComplexDiscovery OÜ

ComplexDiscovery’s mission is to enable clarity for complex decisions by providing independent, data‑driven reporting, research, and commentary that make digital risk, legal technology, and regulatory change more legible for practitioners, policymakers, and business leaders.