For-profit businesses weren’t the focus of the workshop I planned to create,
but there is a lot to learn from the literature on how to shut companies down.
Whetten1980 argued that management science had almost entirely ignored organizational decline
in favour of growth.
WeitzelJonsson1989 extended this into a five-stage model:
blindness (warning signs go unrecognised),
inaction (signs are seen but nothing changes),
faulty action (the wrong remedies are tried),
crisis (resources are depleted and options close off),
and dissolution.
The practical insight is that by the time closure feels urgent,
the organisation has usually been in decline for much longer than anyone acknowledged.
For project teams,
this predicts that the moment when shutdown becomes undeniable is rarely the moment it became inevitable:
the warning signs were there earlier,
and recognising the stage a project is in determines which interventions are still available.
Exit Planning Is a Skill That Takes Time to Develop
Burlingham2014 interviewed entrepreneurs who had closed or sold their businesses
and identified a consistent finding:
the owners who were satisfied with how their exits went were those who had started planning years in advance.
He organises the exit process into four stages
he calls exploration, strategy, execution, and transition,
and shows that skipping the first two because they feel premature is the most common mistake.
Warrillow2011 makes a complementary argument:
a project or business that can only be run by its founders has no transferable value,
and building transferable value requires deliberate, ongoing effort
rather than a last-minute handover document.
Founder Departure Has a Psychology of Its Own
Sonnenfeld1988 studied what happens when leaders leave the organisations they built
and identified four departure archetypes:
Monarchs refuse to leave until forced out and undermine successors.
Generals depart reluctantly, maintain close ties, and often attempt to return.
Ambassadors leave willingly, remain available to successors, and are the most satisfied post-exit.
Governors leave cleanly for new pursuits and disengage entirely.
Which of these a founder becomes is shaped by how central the project is to their identity;
recognising which pattern someone is likely to follow
is useful preparation for the handover conversations.
Financial Distress Has Direct and Indirect Costs
Wruck1990 analysed the cost of financial distress in a large sample of companies
and found that direct costs like legal fees and asset sales at distress prices
were typically modest relative to enterprise value.
On the other hand,
indirect costs like lost customers, departing employees, cancelled contracts, and management distraction
were substantially larger and harder to measure.
The indirect costs arise because distress signals that the organisation
may not be able to fulfil its commitments,
which triggers defensive behaviour that in turn accelerates the decline.
For project teams under funding pressure,
the analog is the chilling effect on contributions when a shutdown is rumoured but not yet confirmed:
the information vacuum does more damage than the honest announcement would have done.
Grief Is a Normal Response to Business Failure
Shepherd2003 and Shepherd2016 found that
the intensity of grief experienced by entrepreneurs after a failed venture
is proportional to the degree of personal identification with the project,
and that it directly interferes with the ability to learn from the experience.
Crucially,
grief recovery is not linear
and is not accelerated by forcing “lessons learned” conversations too early.
Instead,
structured reflection is most productive when it begins after a period of acknowledgement,
not immediately after the loss.
This suggests that
a retrospective held the week a project closes will capture less than one held a month later,
once the initial shock has passed.
The Asset Sale vs. Entity Sale Distinction
When a company closes or transfers its work to another organisation,
the transaction is typically structured either as an asset transfer
(selected assets and obligations are transferred,
but the original legal entity persists or is wound down separately)
or as an entity transfer (the entire legal entity, including unknown liabilities, changes hands).
Marks2022 covers this distinction in detail in the context of private company transactions,
and the same logic applies to project closure:
handing over a GitHub organisation,
a domain name,
and a mailing list archive
is not the same as handing over the project’s governance obligations or data privacy commitments.
Being explicit about exactly what is and isn’t being transferred
protects both the outgoing project team and the incoming maintainers.
Every handover should therefore be treated as an asset transfer with an explicit list of what is included,
rather than assuming that “the project” is a single indivisible thing.
Legal Obligations Survive Shutdown
Business closure triggers a set of legal obligations that are easy to overlook and expensive to ignore.
Creditors,
including employees with unpaid wages,
suppliers with outstanding invoices,
and landlords with remaining lease terms,
are entitled to notice.
In most legal systems
they also have a defined period in which to submit claims before residual assets are distributed.
In addition,
most jurisdictions require formal dissolution filings with the government
and final tax returns covering the period up to the dissolution date;
failure to file typically results in ongoing tax and reporting obligations
even after the organisation has ceased to operate.
Intellectual property assets do not transfer automatically on dissolution
and require explicit assignment or licensing agreements.
Employment notice requirements vary widely across jurisdictions
but typically require advance written notice of redundancies above a threshold size,
with failure to comply entitling affected employees to compensation.
Project leaders must therefore identify jurisdiction-specific equivalents of these obligations
early in the closure process,
not as a final step,
because some of them (such as creditor notification periods) create binding deadlines
that constrain everything else.
Bibliography
Burlingham2014
Bo Burlingham:
Finish Big: How Great Entrepreneurs Exit Their Companies on Top.
Portfolio/Penguin,
2014,
978-1591844976.
Marks2022
Kenneth H. Marks, Christian W. Blees, Michael R. Nall, and Thomas A. Stewart:
Middle Market M&A: Handbook for Advisors, Investors, and Business Owners
(2nd ed.).
Wiley Finance,
2022,
978-1119828105.
Shepherd2003
Dean A. Shepherd:
“Learning from Business Failure: Propositions of Grief Recovery for the Self-Employed.”
Academy of Management Review,
28(2),
2003,
10.5465/amr.2003.9416377.
Shepherd2016
Dean A. Shepherd, Trenton Williams, Marcus Wolfe, and Holger Patzelt:
Learning from Entrepreneurial Failure: Emotions, Cognitions, and Actions.
Cambridge University Press,
2016,
978-1107569836.
Sonnenfeld1988
Jeffrey Sonnenfeld:
The Hero’s Farewell: What Happens When CEOs Retire.
Oxford University Press,
1988,
978-0195065831.
Ucbasaran2013
Deniz Ucbasaran, Dean A. Shepherd, Andy Lockett, and S. John Lyon:
“Life After Business Failure: The Process and Consequences of Business Failure for Entrepreneurs.”
Journal of Management,
39(1),
2013,
10.1177/0149206312457823.
Warrillow2011
John Warrillow:
Built to Sell: Creating a Business That Can Thrive Without You.
Portfolio,
2011,
978-1591843979.
WeitzelJonsson1989
William Weitzel and Ellen Jonsson:
“Decline in Organizations: A Literature Integration and Extension.”
Administrative Science Quarterly,
34(1),
1989,
10.2307/2392987.
Whetten1980
David A. Whetten:
“Organizational Decline: A Neglected Topic in Organizational Science.”
Academy of Management Review,
5(4),
1980,
10.5465/amr.1980.4288962.
Wruck1990
Karen Hopper Wruck:
“Financial Distress, Reorganization, and Organizational Efficiency.”
Journal of Financial Economics,
27(2),
1990,
10.1016/0304-405X(90)90063-6.
The preceding posts assume that project members have time to prepare for closure,
but reality is rarely that accommodating.
When closure arrives,
the question is usually not how to execute a plan but how to act usefully in its absence.
Crisis response research has studied exactly this situation across many domains,
and its findings offer practical guidance for project participants who have no script to follow.
When people describe expert decision-making they tend to imagine a deliberate weighing of options,
but Klein1998 tells a different story.
Under time pressure, experts do not compare options:
they recognize a situation as an instance of a familiar type
and act on the first workable course of action that comes to mind.
Klein called this recognition-primed decision making because the decision is primed by experience,
not constructed from first principles.
The implication for project teams caught in unplanned closure is that
the experience they have dealing with emergencies in other domains
is more actionable than they realize.
Asking “has anyone seen something like this before?” is not a distraction:
it is a direct way to activate usable tacit knowledge under pressure.
Psychological First Aid
Hobfoll2007 identified five empirically-supported intervention principles
that should guide intervention and prevention efforts in the early stages of disasters:
promoting a sense of safety,
calming,
self-efficacy and collective efficacy,
connectedness,
and hope.
These were distilled from decades of disaster mental health research
and became the basis of the WHO’s Psychological First Aid field guide
WHO2026.
The practical translation for project teams is that
the first conversations after an unplanned closure announcement
should not be about logistics.
Instead,
project leaders should confirm that people are not in immediate danger,
lower the emotional temperature by acknowledging what has happened,
remind people that they have handled difficult things before,
connect people with one another rather than leaving them to absorb the news alone,
and identify whatever is genuinely uncertain versus irreversibly decided.
Attention Narrowing and the Value of Lists
Attention narrows when people are under acute stress.
Easterbrook1959 research showed that this reduces the range of cues a person notices,
and subsequent work has confirmed that
cognitive stress specifically impairs
the ability to hold multiple competing priorities in mind simultaneously Starcke2012.
The practical effect for a team facing project closure is that they will focus intensely
on the most visible problem in front of them,
such securing the code repository or notifying funders,
while entirely failing to think of categories of action
that would have been obvious under calmer circumstances.
This is not incompetence: it is a predictable feature of how the brain responds to pressure.
Similarly,
Gawande2009 found that the most common failures in surgery
were not failures of skill but failures to perform steps the surgeon knew needed to be done.
The intervention that worked was not additional training but creating a checklist
that prevented the surgeon’s attention from being displaced by the immediate demands of the situation.
The same principle applies directly to unplanned project closure.
Spend the first few minutes of the crisis writing down every category of thing that must be addressed
(e.g., access credentials, data backups, stakeholder notifications, community communication, legal obligations)
before attempting to resolve any of them.
Teams that do this will cover far more ground than ones that begins triaging or acting immediately.
The list does not need to be complete or correctly ordered:
it just needs to exist,
so that work displaced by urgent action can be recovered rather than forgotten.
Convergence and Emergent Groups
Foundational research on disaster response shows that
people converge on a crisis site without being told to Fritz1957.
Dynes1970 extended this to create a typology of groups that form in emergencies:
established organizations following their routines
organizations that extend their routines to cover the new situation
ad hoc groups that form specifically to respond
emergent groups that have no prior existence at all
In software and research projects,
community members who have no formal role often step forward during unplanned closures:
one person maintains an unofficial mirror,
another documents the shutdown in a public post,
while a third starts an email thread for those affected.
Project leads should recognize this convergence as a resource rather than a complication
and find ways to coordinate with it rather than ignoring or suppressing it.
The people converging are doing so because the project mattered to them,
and they are likely to carry it forward if they are given the minimum viable information to do so.
Street-Level Discretion
Lipsky1980 found that frontline workers under pressure
routinely depart from formal rules when resources are insufficient to apply them as written.
They do not do this casually;
rather,
they do it because applying the rule mechanically would produce worse outcomes than exercising discretion.
This is called street-level bureaucracy:
the real policy of an organization is made not in headquarters
but in the moment-to-moment decisions of its frontline workers.
During unplanned closure,
the formal procedures for data preservation,
access management,
or stakeholder notification are often inapplicable
because there is not enough time,
the people responsible are unavailable,
or the infrastructure has already been shut down.
In this situation,
the people present should not wait for authorization they cannot obtain
before taking actions they can see are necessary.
Copying data,
setting up a forwarding address,
or sending a brief public notice are all within the discretion of the people present.
Waiting for formal approval when formal approval is not available is itself a decision,
and usually a worse one.
Crisis Communication Without a Playbook
Coombs2019 studied how organizations respond to crises they did not anticipate
and identified three core communication errors:
Silence: saying nothing while affected stakeholders fill the vacuum with rumour.
Denial: asserting that the crisis is not as serious as it appears.
Diminishment: acknowledging the crisis but minimizing its severity.
All three increase reputational damage and community harm.
The research consistently shows that rapid, honest, minimal communication is better,
even when (or especially when) the organization does not yet know what it will do next.
A single paragraph in the project’s primary channel saying
“Here is what we know,
here is what we do not yet know,
and here is when we expect to have more information”
is not a plan,
but can prevent the harms of silence.
Emotion-Focused Coping
Lazarus1984 distinguishes two broad categories of coping:
problem-focused coping,
which addresses the situation itself,
and emotion-focused coping,
which addresses the person’s internal response to the situation.
Their research shows that when people perceive a stressor as controllable,
they adopt problem-focused strategies;
when they perceive it as uncontrollable, they adopt emotion-focused ones.
Unplanned project closure typically combines both.
As a consequence,
team members will respond to the same event differently depending on whether they feel any agency over it,
and both responses are appropriate in their own domain.
The person who immediately starts triaging data is not more competent than
the person who needs an hour to absorb what has happened:
they are applying different coping modes to the same situation.
Managing an unplanned closure well means making space for both responses simultaneously.
Tasks that require immediate action should be handled by whoever is in a problem-focused state.
Tasks that require social coordination should not wait until the logistics are resolved,
because people who feel unacknowledged withdraw from the collective effort.
The two streams of work need to run in parallel.
Bibliography
Coombs2019
W. Timothy Coombs:
Ongoing Crisis Communication: Planning, Managing, and Responding
(6th ed.).
SAGE Publications,
2021,
978-1071816646.
Easterbrook1959
J.A. Easterbrook:
“The Effect of Emotion on Cue Utilization and the Organization of Behavior.”
Psychological Review,
66(3),
1959,
10.1037/h0047707.
Dynes1970
Russell R. Dynes:
Organized Behavior in Disaster.
D.C. Heath,
1970.
Fritz1957
Charles E. Fritz and J.H. Mathewson:
Convergence Behavior in Disasters: A Problem in Social Control.
National Academy of Sciences / National Research Council,
1957.
Hobfoll2007
Stevan E. Hobfoll, Patricia Watson, Carl C. Bell, Richard A. Bryant,
Melissa J. Brymer, Matthew J. Friedman, Merle Friedman,
Berthold P.R. Gersons, Joop T.V.M. de Jong, Christopher M. Layne,
Shaul Maguen, Yuval Neria, Ann E. Norwood, Robert S. Pynoos,
Dori Reissman, Josef I. Ruzek, Arieh Y. Shalev, Zahava Solomon,
Alan M. Steinberg, and Robert J. Ursano:
“Five Essential Elements of Immediate and Mid-Term Mass Trauma Intervention:
Empirical Evidence.”
Psychiatry: Interpersonal and Biological Processes,
70(4),
2007,
10.1521/psyc.2007.70.4.283.
Gawande2009
Atul Gawande:
The Checklist Manifesto: How to Get Things Right.
Metropolitan Books,
2009,
978-0805091748.
Klein1998
Gary Klein:
Sources of Power: How People Make Decisions.
MIT Press,
1998,
978-0262611466.
Lazarus1984
Richard S. Lazarus and Susan Folkman:
Stress, Appraisal, and Coping.
Springer,
1984,
978-0826141910.
Lipsky1980
Michael Lipsky:
Street-Level Bureaucracy: Dilemmas of the Individual in Public Services.
Russell Sage Foundation,
1980,
978-0871545268.
Starcke2012
Katja Starcke and Matthias Brand:
“Decision Making Under Stress: A Selective Review.”
Neuroscience and Biobehavioral Reviews,
36(4),
2012,
10.1016/j.neubiorev.2012.02.003.
WHO2026
World Health Organization:
“Psychological first aid: Facilitator’s manual for orienting field workers.”
https://www.who.int/publications/i/item/psychological-first-aid,
2013,
viewed 2026-04-08.
Last year I started building a workshop on how to shut projects down.
I put it on hold in the fall,
both because I was finding it difficult to figure out what to say
and because working on something like that after being laid off felt a bit unhealthy.
A couple of recent discussions have revived my interest,
though,
so here are some notes that I hope one day will find a home.
Vaughan1996 showed that the engineers involved in the Challenger disaster
did not make a single catastrophic decision
but instead made a long sequence of small decisions in which signals of danger were repeatedly noticed,
reinterpreted as acceptable,
and filed away until deviation from safety standards had become the practical norm.
The parallel for software and research projects is that
teams approaching closure often carry a backlog of things that everyone knows are wrong
but that have never been formally acknowledged.
A deliberate closure process should include an explicit audit of normalized deviations
that the team accepted because stopping to fix them felt too costly,
both to honestly document the project’s actual state
and to avoid passing silent assumptions on to anyone who inherits the work.
In practical terms,
a shutdown retrospective is most valuable not when it celebrates successes
but when it names the things the team learned to live with.
Fantasy Documents and the Rhetoric of Preparedness
Clarke1999 describes how organizations routinely produce disaster response plans
that have no realistic chance of being executed.
Clarke calls these “fantasy documents”:
their purpose is to reassure regulators, funders, and the public that someone is in charge,
despite bearing no relationship to what would actually happen under stress.
For software and research projects,
this maps onto the shutdown plan that exists in a drawer but has never been tested,
or to a HANDOVER.md written in an afternoon that no one outside the core team has ever read.
A closure plan should be treated as a drill script, not a policy document:
teams should periodically walk through it step by step
and ask whether each action is actually achievable with the people available.
Tight Coupling and Normal Accidents
Tight coupling means that failure in one part of a system propagates rapidly through others
with little time for intervention.
Perrow1999 argues that in complex systems where components are tightly coupled,
accidents are not aberrations but predictable outcomes of the system’s design.
Modern software and research infrastructure is often tightly coupled
in ways that only become visible during shutdown:
an account that pays for ten services,
or a single personal login that controls a domain.
These tight couplings mean that an unexpected departure or account closure
cascades into a complete loss of access before anyone has time to respond.
The solution is to treat any single point of control that cannot survive a 48-hour unavailability as a bug.
That said,
Singer2023
contrasts “blame the individual” with “blame the system”,
and shows that companies push the latter to absolve themselves of responsibility.
The Survival Arc
Ripley2008 synthesises research on how people behave when disaster strikes
and puts forward a three-stage survival arc:
Denial: refusal to believe that the situation is as serious as evidence suggests.
Deliberation: a period of paralysed uncertainty about what to do.
Decision: action, which may or may not be correct.
Research on building evacuations, shipwrecks, and plane crashes consistently shows that
the denial phase consumes far more time than people expect,
while thinking through the scenario before it happens dramatically shortens it.
For project closure,
this predicts that the period between a credible signal that a project is in trouble
and the first concrete closure action will be longer than planners assume,
particularly under abrupt conditions.
Rehearsal converts a novel emergency into a recognised pattern.
Warning Signs
While it is not directly relevant to project closure,
Palazzo2025 argues that there are patterns in the dynamics of corporate scandals.
Would recognizing them give people early warning that projects were going to fail?
rigid ideology
toxic leadership
manipulative language
perceived unfairness
dangerous groups
slippery slope
The Collapse of Sensemaking
Weick2007 studied organisations that operate reliably in high-stakes, high-complexity environments
and identified five shared practices they call mindful organizing.
Two of these are relevant to project closure:
sensitivity to operations
(i.e., maintaining an accurate picture of what is actually happening, not what plans say is happening)
and commitment to resilience (the capacity to improvise when standard procedures fail).
Earlier, Weick1993 introduced the idea of a cosmology episode:
a moment when the world stops making sense and actors freeze.
These can occur during abrupt project closures
when the governance structures that normally coordinate work suddenly disappear.
One way to deal with this is to make sure that teams have fallbacks
such as personal email addresses or a shared document that doesn’t depend on organizational infrastructure.
The twin risks of this are that only insiders will know these unofficial documents exist,
and that they will be out of date by the time they’re needed.
Anticipatory Mourning and the Grief of Planned Endings
Rando2000 studied anticipatory mourning:
grief that begins before an expected loss, not after it.
She distinguishes this from simple “grief in advance”:
it involves simultaneous processing of past losses (what the project once was),
present losses (what it is becoming),
and future losses (what it will never be),
and it is cognitively and emotionally more demanding than grief after the fact
because the person must continue to function while still engaged in the thing they are losing.
For project teams,
planned shutdown triggers this:
team members must continue doing the work of closing down
while processing the loss of their work relationships and the future they had imagined.
A practical recommendation is to name this dynamic explicitly,
because people who understand that their feelings are real are better able to manage them
and to support one another.
Bibliography
Clarke1999
Lee Clarke:
Mission Improbable: Using Fantasy Documents to Tame Disaster.
University of Chicago Press,
1999,
978-0226109411.
Palazzo2025
Guido Palazzo and Ulrich Hoffrage:
The Dark Pattern: The Hidden Dynamics of Corporate Scandals.
PublicAffairs,
2025,
978-1541705302.
Perrow1999
Charles Perrow
Normal Accidents: Living with High-Risk Technologies.
Princeton University Press,
1999,
978-0691004129.
Rando2000
Therese A. Rando (ed.):
Clinical Dimensions of Anticipatory Mourning: Theory and Practice in Working with the Dying, Their Loved Ones, and Their Caregivers.
Research Press,
2000,
978-0878223800.
Ripley2008
Amanda Ripley:
The Unthinkable: Who Survives When Disaster Strikes—and Why.
Crown Publishers,
2008,
978-0307352897.
Singer2023
Jessie Singer:
There Are No Accidents: The Deadly Rise of Injury and Disaster—Who Profits and Who Pays the Price.
Simon & Schuster,
2023,
978-1982129668.
Vaughan1996
Diane Vaughan:
The Challenger Launch Decision: Risky Technology, Culture, and Deviance at NASA.
University of Chicago Press,
1996,
978-0226851754.
Weick1993
Karl E. Weick:
“The Collapse of Sensemaking in Organizations: The Mann Gulch Disaster.”
Administrative Science Quarterly,
38(4),
1993,
10.2307/2393339.
Weick2007
Karl E. Weick and Kathleen M. Sutcliffe
Managing the Unexpected: Resilient Performance in an Age of Uncertainty
(2nd ed.).
Jossey-Bass,
2007,
978-0787996499.
Eight years on, I still miss my brother almost every day.
I haven’t read much about medical assistance in dying (MAID) since he took advantage of it,
but I still think there are lessons there for ending projects.
As with the first post in this series, though,
I worry about how much of this people will actually do before it’s needed,
particularly when they’re already under stress.
An advance directive (sometimes called a living will) is a document someone writes
while they are still healthy and competent,
specifying what care they do and don’t want if they become unable to speak for themselves.
The direct parallel for a software or research project is something prepared during normal operations
that specifies the conditions under which the project should be wound down,
who has the authority to make that call,
what assets must be preserved and how,
and what obligations must be honoured.
Towey2026 extends this idea beyond purely medical instructions
to cover personal, emotional, and relational matters,
which maps well onto the non-technical dimensions of project closure.
The most important thing is to recognize that this document needs to be written before it is needed:
once a project is under stress,
cognitive load and interpersonal conflict make clear-headed decision-making much harder,
and governance discussions that should have happened earlier are forced into the worst possible moment.
Total Pain and the Hidden Costs of Closure
Dame Cicely Saunders, who founded the modern hospice movement,
introduced the concept of total pain in the 1960s
to describe how a dying patient’s suffering is never purely physical:
it has physical, psychological, social, and spiritual dimensions that interact and amplify one another Saunders1963.
Project closure similarly involves more than logistics:
team members experience professional identity loss,
grief over unfinished work,
anxiety about career prospects,
and sometimes guilt or anger that can fracture long-standing relationships.
Treating project closure as a purely administrative exercise leaves people worse off
and makes it harder for the broader community to learn from the experience.
Goals-of-Care Conversations
Gawande2014 argues that clinicians systematically fail patients by focusing on what medicine can do
rather than asking what the patient actually wants.
He advocates a structured approach to hard conversations:
establish a shared understanding of the current situation,
make the other person’s values and priorities explicit,
and then negotiate a plan that fits those goals rather than being driven solely by what is technically possible.
Similarly,
when a project is ending,
its leaders must talk with funders, institutional sponsors, and the team
about what matters most:
not just what data can be preserved,
but what is the least bad outcome for each stakeholder,
and what trade-offs each party is willing to make.
Dignity Therapy and Legacy Documents
Chochinov2005 and Chochinov2012 advocate dignity therapy
as a brief, structured intervention for people near the end of life.
The central technique is a guided life-review interview
in which the person reflects on what has mattered most,
what they are proud of,
and what they most want future generations to know and remember.
The session is recorded, transcribed, and edited into a permanent legacy document
that extends the patient’s influence beyond their death.
A similar structured exit interview or retrospective at the end of a project
is more than a conventional “lessons learned” report
because it is personal, narrative, and explicitly oriented toward what should survive rather than what went wrong.
Continuing Bonds
The continuing bonds theory of bereavement Klass1996
challenged the idea that healthy mourning requires letting go of the deceased and severing the relationship.
Instead,
it argues that bereaved people who maintain an ongoing internal relationship with the dead
through story-telling, physical objects, or integration of the deceased’s values into their own identity
are not pathological:
they are exhibiting a normal and healthy form of grief.
The lesson for project closure is that closing a project is not the same as erasing it,
and that participants don’t need to move on in a way that severs their connection to the work.
Design decisions, published papers, open-source code, and the skills people developed all continue to exist and to matter.
Framing closure around what continues,
rather than only around what ends,
reduces distress and gives project leaders a vocabulary for talking about archiving, citation, and attribution
not as bureaucratic compliance but as genuine acts of preservation.
Bibliography
Chochinov2005
Harvey Max Chochinov, Thomas Hack, Thomas Hassard, Linda J. Kristjanson, Susan McClement, and Mike Harlos
“Dignity therapy: A novel psychotherapeutic intervention for patients near the end of life.”
Journal of Clinical Oncology,
23(24),
2005,
10.1200/JCO.2005.08.391.
Chochinov2012
Harvey Max Chochinov:
Dignity Therapy: Final Words for Final Days.
Oxford University Press,
2012,
978-0195176216.
Clark1999
David Clark:
“Total pain, disciplinary power and the body in the work of Cicely Saunders, 1958-1967.”
Social Science and Medicine,
49(6),
1999,
10.1016/S0277-9536(99)00098-2.
Gawande2014
Atul Gawande:
Being Mortal: Medicine and What Matters in the End.
Metropolitan Books / Henry Holt,
2014,
978-0805095159.
Klass1996
Dennis Klass, Phyllis R. Silverman, and Steven Nickman (eds.):
Continuing Bonds: New Understandings of Grief.
Taylor and Francis / Routledge,
1996,
978-1560323396.
Saunders1963
Cicely Saunders:
“The treatment of intractable pain in terminal cancer.”
Proceedings of the Royal Society of Medicine,
56(3),
pp. 195-197,
1963.
Towey2026
Jim Towey:
Five Wishes.
Aging with Dignity,
https://www.fivewishes.org/,
viewed 2026-04-08.
The project I’d really like to be working on is a tower support game
(like tower defense, but the player is trying to help travelers reach their destinationn instead of kill them).
If you’re interested in collaborating on that,
please reach out.
As I wrote a couple of days ago,
Sajaniemi et al’s work on roles of variables
identified and named ten patterns in the way variables are used in novice programs.
These roles aren’t just a way to analyze code.
They’re also useful for explaining code,
because they provide a vocabulary to capture intent
in a way that’s complementary to type information like int or Person.
Role
Description
Fixed value
Initialized once and not changed thereafter.
Stepper
Steps through a systematic, predictable succession of values (e.g., a loop counter).
Walker
Traverses a data structure element by element, where each new value depends on the previous position.
Follower
Always receives the previous value of some other variable (e.g., prev = current before current is updated).
Most-recent holder
Holds the latest value encountered in an unpredictable series, typically successive inputs.
Most-wanted holder
Holds the best value found so far (e.g., maximum or minimum).
Gatherer
Accumulates the combined effect of individual values (e.g., running sum or count).
One-way flag
A two-valued variable that, once changed from its initial value, never reverts (e.g., an error flag set to True and never reset).
Temporary
Holds a value for a very short time only, most commonly in a swap operation.
Organizer
A collection used only to rearrange its elements (e.g., sorted in place), never to add or remove them.
Container
A collection whose elements are added and/or removed during the computation.
It occurred to me during a workshop this week that
variable roles might help when we are using LLMs in teaching.
In order to understand a piece of software,
I need to understand the purpose of each of its parts.
Pattern names like Observer or Singleton help me do this
because they compress complex chunks of code down to single concepts,
which helps prevent cognitive overload.
However,
the classic design patterns are still too big for novices;
would Sajaniemi’s roles be more helpful?
I.e.,
can we prompt the LLM to explain things in terms of these roles
while simultaneously teaching newcomers to think in terms of them,
and if so,
does it help?
And if the answers are “yes”,
are there other roles that Sajaniemi and his colleagues didn’t identify
that are worth adding to the table above?
(I work with data scientists
so the first that comes to mind is “Dataframe”,
but that’s not a role: that’s a data type.)
One way to think about roles is that types tell you about the variable’s state at rest
while roles tell you about the variable’s state in motion.
This duality is similar to the one Kerievsky described
in his under-appreciated book Refactoring to Patterns,
which explored the duality between design patterns (code at rest)
and refactorings (code in motion).
In order to explore these ideas,
I spent about three and a half hours prompting Claude to build jorma,
a small Python package that uses static and dynamic analysis
to identify variables’ roles.
At the time of writing it is about 1000 lines of Python (not including tests)
and is able to process 145 source files taken from Software Design by Example.
I made a few small edits to the code manually,
but almost all of what’s in src and tests was written by Claude.
I think it would have taken me four or five days of full-time work and frustration
to get this far on my own,
which means that I probably wouldn’t have done it.
That realization is what led to this post’s title:
if a computer is a bicycle for the mind,
then LLMs are like e-bikes.
They let a lot of people go distances and tackle hills that they couldn’t before,
and they’re better for all of us than cars,
but they’re a menace to both pedestrians and traditional cyclists,
more harmful to the environment than what they’re replacing,
and have given companies yet another way to hollow out local businesses.
(Neighborhood restaurants know that cheap delivery services are killing them slowly,
but if they don’t play,
they’ll just die more quickly.)
I wish I could write a tidy conclusion for this post,
but I don’t have one.
Experiments like jorma have convinced me that
telling people not to use LLMs
is going to be just as ineffective as telling teenagers not to have sex or use social media.
I hope that this week’s court decisions to hold Meta accountable for the harm it has caused
is a first step toward dealing with cognitive pollution in general,
for the same reason that I hope Toronto and other places will figure out
how to make e-bikes a normal part of everyday life.
Until then,
if you’d like to help me improve jorma,
please try it out or get in touch.
RSECon26 is happening in Sheffield in September.
After spending two days talking to people about LLMs and research software engineering,
I think the conference will be a golden opportunity to find out
how people are currently using LLMs in their work.
Invite people to take part in a 90-minute session
(which, once they’re set up, will give you about an hour of useful time).
Give them two small problems to solve:
one that requires writing code from scratch,
one in which they need to debug existing code.
(I’d use data ingestion and cleanup problems,
because I think they’ll be most familiar to most people.)
Have them record their screens while they work.
They can use whatever tools they want:
in particular,
they can use or not use LLMs when and as they normally would.
Analyzing those recordings will give us a baseline reference
against which to to compare future runs of this experiment.
It will also help us identify places where some people are more or less advanced than others,
which will give us some idea of what to teach
and a reason for people to pay attention.
Most importantly,
I think it will give people in the RSE community something to look forward to
and a reason to believe that people they trust are trying to figure out what should happen next.