Hunger doesn’t clock out
The people who take the work home are how a company learns the frontier. How to find them and keep them
9 min read
TL;DR: A small number of people in any company take a piece of work home. They spend evenings and weekends on it, reskill to keep pace with it, and push the edge of what is known forward by a little. At the frontier — AI right now — they are the only way the company learns, because there is no course to send people on and no expert to hire. This blog has visited that territory three times from the individual’s side. This is the view from the other side of the desk: how to recognise these people without mistaking spare hours for appetite, why the performance system will penalise them without meaning to, and why recognition after the fact is the one reward that cannot be gamed.
Work that is already defined
Most of the time, and for most of us, the hours we spend at work go into work that is already defined. The shape changes as we go up — from tasks with a definition of done to two-year objectives that need a strategy and a mission statement — but it is still work that is known, and we are paid for it.
Sometimes, and for a small number of people, that equation changes. A piece of work gets under the skin. They take it home. They spend evenings and weekends on it, read up and reskill to keep pace with it, and push the edge of what is known forward by a small amount. Three parties gain: the individual, who is better at the end of it than at the start; the company, which now holds a capability it never budgeted for; and occasionally the rest of us.
Linux is the standard example. On August 25, 1991, a Finnish student posted to the comp.os.minix newsgroup:
Hello everybody out there using minix — I’m doing a (free) operating system (just a hobby, won’t be big and professional like gnu) for 386(486) AT clones. This has been brewing since april, and is starting to get ready.
All 500 of the world’s fastest supercomputers have run Linux since November 2017, and the June 2026 list is no exception.
The question for a manager, or for whoever is spearheading an innovation initiative, is how to find these people accurately, back them, and build the ecosystem around them so that the individual, the initiative, the company and sometimes the wider field all come out ahead.
We have visited this before
My 2024 post argued that the hours on either side of the working day were the only ones that were properly yours. A 2025 post made the case that those hours are also where careers turn, if they go into building things the job never asked for. A third introduced the unwritten résumé: the side quest, the small tool built for fun, the hallway conversation about your work in a room you were not standing in.
All three spoke to the individual. I have sat on both sides of that desk: the engineer spending a weekend on something the job never asked for, and the manager who inherited a Monday-morning prototype, put it in a deck by Wednesday, and never asked whose weekend it had cost. That second one has happened more times than I would like to count. Writing about taking the work home covers half of a transaction. The other half is whether the place you bring it back to knows what it is looking at.
Burnout is real, and “opt-in” loses its moral shine the moment it starts to smell like social coercion.
The unbudgeted department
Most companies carry a workforce that does not appear on org charts: people running a project that was never approved, learning a tool that was never procured, or keeping alive a thing that was formally cancelled. Technologists call it a skunkworks, after the Lockheed division Kelly Johnson ran from 1943. The third of his fourteen rules for it:
The number of people having any connection with the project must be restricted in an almost vicious manner.
Johnson’s version had a budget and a sponsor. Management researchers have a separate name for the version that has neither — bootlegging — and Paola Criscuolo, Ammon Salter and Anne ter Wal open their study of it with a sentence most engineers will recognise:
To develop innovations in large, mature organizations, individuals often have to resort to underground, “bootleg” research and development (R&D) activities that have no formal organizational support.
They studied it across a large R&D organisation and found that the people who bootleg are the ones with the strongest innovation records. The effect is larger in units where colleagues are doing the same, and it reverses when the company is tightening its process. Hunger produces results in proportion to the room it is given.
Get Kundansen’s stories in your inbox
Join Medium for free to get updates from this writer.
A company at a genuinely new frontier has few things to buy. Ten years of experience in a three-year-old field is unavailable at any price. A course will be stale before it ends. A vendor moves the risk of failure off the company’s books, and is learning at the same rate while charging for the privilege. What the company has is a handful of employees who have been reading papers on the train, running models on their own hardware, and building small things at the weekend that mostly did not work — until one did.
That is where frontier capability comes from right now, in tech and tech-adjacent organisations everywhere. The strategy deck describes it as an initiative. The people who built it know it as a series of evenings.
MIT’s Project NANDA put a number on the gap in 2025. Of the formal generative-AI initiatives it reviewed, 95% showed no measurable effect on profit and loss. In the same year, employees at more than 90% of companies were using personal chatbot accounts for daily work, while only 40% of companies had an official subscription. The report calls it a shadow AI economy. It is the unbudgeted department, counted.
Scott Berkun’s seventh Myth of Innovation is that your boss knows more about innovation than you. His subtitle for the chapter is don’t confuse power with creativity. A manager’s authority is over decisions, not over where the ideas come from, and the ideas come from wherever the work is being done. The job is to make sure that when one arrives there is somewhere for it to go.
What to look for
Lencioni’s definition of hunger is the one most managers carry around, whether or not they have read him:
Hungry people are always looking for more. More things to do. More to learn. More responsibility to take on. Hungry people almost never have to be pushed by a manager to work harder because they are self-motivated and diligent.
It is accurate, and it says nothing about what to observe.
Look first for what already exists without permission. Hunger shows up before it is looked for: the internal tool with forty users and no owner, the notebook that turns up in three other people’s decks, the AI skill that dozens of colleagues use because it was passed around in one-to-one chats. What the appetite has already produced is a better interview than any question about it. Gifford Pinchot, who coined the word intrapreneur in 1985, wrote a list of commandments for the role. Two of them:
Work underground as long as you can.
Come to work each day willing to be fired.
The work that survives without air cover is the work worth funding.
Look for the champion as well as the builder. Berkun’s eighth myth is that the best ideas win; his subtitle for the chapter is winning isn’t based on how good the idea is. The ones that win had a champion who kept explaining them after the third no. Back the one still explaining when the explaining has stopped being fun. Persistence after the reward stops is the part of hunger that cannot be faked with a weekend and a template.
Look for the one who can stop. Robert Vallerand’s research splits passion into two kinds that look identical from the outside for about six months. Harmonious passion is chosen and sits alongside the rest of a life; the engineer can set it down on Friday and pick it up again for years. Obsessive passion is driven by pressure or by self-worth, crowds out the rest of a life, and ends in exhaustion. Back the one who still shows up with their boundaries intact.
Look for intensity per hour, not for hours. If every hungry person you have identified looks the same — early in their career, no dependants, in the same time zone as the optional call — then you have been reading availability and calling it appetite. The parent doing forty-five focused minutes at lunch is hungry. The engineer in Bengaluru who cannot join the 10 PM demo night but has rewritten the evaluation suite by morning is hungry. The earlier post listed those constraints as reasons a volunteer might not show up. Each one is a price the manager can lower — move the demo, make the channel async, protect an hour in the day:
Life-stage constraints: parenting, elder care, a side degree — time is fragile. A monthly demo night that starts at 7 PM can feel like a tax on sleep. A global workforce makes it worse: the call time that suits New York lands in Bengaluru at 11 PM.
Why the system will not do it for you
Diane Bergeron and colleagues titled their study “The Cost of Being a Good Citizen.” Across 3,680 professionals in a firm that rewarded measured outcomes, time spent on discretionary work was associated with slower promotion and lower salary growth. The mechanism is arithmetic: the hours came out of the work that was being counted, and the performance system does not see the weekend.
Recognition therefore has to come from a manager, because the process cannot see the thing being recognised. Goodhart’s law says a measure that becomes a target stops being a good measure — announce that side projects count toward promotion and you will have side projects by the following quarter, none of which their owners took home because they wanted to. Recognition after the fact, for work that was never asked for, is the one reward Goodhart cannot reach: it was never a target, so it was never gamed.
The missing half
My 2024 post describes three months of treating both windows as pure throughput — courses in the morning, projects in the evening, no unstructured time — and hitting a wall: reading without retaining, writing code that worked and learning nothing from it. It was published as a caution to the individual. It never made the connection to the manager who would take a colleague’s Monday-morning prototype into a Wednesday deck without asking what it had cost, or noticing that its builder had gone quiet by Friday.
If you want hungry people, stop asking how to spot them. Ask whether your work produces them — whether the thing brought back on Monday has somewhere to go, whether its builder is named when it is shown, and whether the price of showing you what they have is one that the parent, the carer and the colleague in another time zone can also afford to pay. The frontier is being explored on evenings the company never paid for. The least it can do is know whose.
Further reading: Scott Berkun’s The Myths of Innovation (2007, revised 2010) is the clearest short argument that ideas arrive from wherever the work is being done and win through champions rather than merit — the seventh and eighth of his ten myths carry most of this piece’s weight; his The Year Without Pants (2013) is the inside account of Automattic, a company built on open-source volunteer habits with almost no conventional reporting lines. Paola Criscuolo, Ammon Salter and Anne ter Wal’s “Going Underground: Bootlegging and Individual Innovative Performance” (Organization Science, 2014) is the peer-reviewed study of unofficial project work inside corporate R&D, building on Peter Augsdörfer’s earlier “Bootlegging and Path Dependency” (Research Policy, 2005). Gifford Pinchot’s Intrapreneuring (1985) coined the term and the commandments, including the instruction to come to work each day willing to be fired. Diane Bergeron and colleagues’ “Organizational Citizenship Behavior and Career Outcomes: The Cost of Being a Good Citizen” (Journal of Management, 2013) is the source for the promotion penalty on discretionary effort in outcome-rewarded environments, and Mark Bolino and William Turnley’s “The Personal Costs of Citizenship Behavior” (Journal of Applied Psychology, 2005) documents the overload and work-family conflict that come with it. Robert Vallerand’s dualistic model of passion, developed from 2003 onward, separates harmonious from obsessive passion and is the basis for looking for the one who can stop. Patrick Lencioni’s The Ideal Team Player (2016) is where most managers’ working definition of hunger comes from, and it is worth reading in full to see how little the book says about how to observe it. MIT Project NANDA’s The GenAI Divide: State of AI in Business 2025 (Challapally, Pease, Raskar and colleagues, July 2025) is the source for the 95% figure and the shadow AI economy; note that the same report finds vendor partnerships outperforming internal builds, so the shadow economy is evidence of where capability sits, not a claim about where the successful 5% came from.
