Relay Week

AI is accelerating, public companies are doing layoffs, and many AI startups are asking employees to work 6 days a week. I hold out hope the productivity of AI will manifest into material wealth and leasure time. There is plenty of evidence for AI creating wealth so I will focus on leisure time.

Textile Mills

There is historic precedence for extreme work schedules during a time of automation. In the north of England starting in 1771 a cotton mill in Derbyshire is credited as the first modern factory. The productivity and cost of water and steam equipment gave an advantage to mills that operated for more hours. The employees worked 12 to 14 hour days, 6 days a week, taking Sunday off. The work culture went on that way for multiple generations.

Full Time

The concept of full time work is relatively new. In the preindustrial era laborers worked in the seasonal daylight hours and artisans fulfilled orders. The factory model of the industrial revolution was the start of time based shifts. Reform started in England with the Ten Hours Act in 1847 to limit the hours of women and teenagers.

Robert Owen in Scottland came up with the idea for an 8-hour work day. His proposal was "eight hours labour, eight hours recreation, eight hours rest" in the 1810s. I don't have a catchy tagline yet for the relay week. The idea was picked up by unions in the U.S. in the 1860s and 8-hour days became the standard for federal employees.

The next wave of reform came in the 1920s when the Ford Motor Company adopted a 5-day schedule. The rationale was interesting. Henry Ford wanted to give factory workers more rest and leisure time. He belived it would reduce mistakes on the assembly line, help with retention, and boost consumer spending on Saturdays. It seems reasonable that more leisure time would lead to more consumption today as well.

The New Deal under Roosevelt included the Fair Labor Standards Act of 1938 which introduced the minimum wage and overtime pay. It started at 44 hours per week and reduced to 40 hours. The legislation does not apply to executives, administrators, and professionals.

Full time employment offered today is nonexempt with predictable 40-hour weeks, or exempt with unpredictable 40-plus hour weeks. It has been that way for over 85 years.

Steering Committee

In the last year software engineering has evolved from editing files to generating patches. Most people agree the same level of automation is coming soon for more occupations. Individual contributors are shifting from doing work to supervising machine processes.

The transition from doing to supervising changes the individual scope. Instead of focusing on execution for a single project, employees can supervise multiple projects. An overlooked aspect in predictions on the ‘future of work’ is that right now in the ‘present of work’ high-level work is collective, not individual. Strategic planning, research peer review, user interface design critiques, software architecture design reviews, venture investment partner votes.

It follows that if machines do the work, employees do high level work, team structure will adapt from directors, managers, and individual contributors into small groups of high taste reviewers and decision makers. Instead of an operating department you have a steering committee. The group designs the environment in which the machine processes operate, starting new processes, reviewing and approving milestones, and intervening if necessary.

Project Affinity

In the soon-to-be past of software engineering context was king. The ramp to become productive on a project for low level work was expensive. Managers went to great lengths to maintain stable project assignments to avoid thrash. The risk of duplicate ramp effort and errors made handing off projects a last resort.

In the soon-to-be future of software engineering the cost of handing off projects will be less dramatic. The agent harness will have access to meeting notes, past decisions, project deadlines, the current state of work. The employees will be regularly context switching to make progress on many projects at once. It might be ok to reassign a project to the most relevant and available steering committee at different stages. It might be disorienting to jump into and out of projects, but perhaps not any more so than the baseline of supervising lots of projects.

The other benefit of single assigment is accountability. The responsibility to drive a project forward and complete the project is unambiguous if there is a single owner. The simplicity was valuable in the era of artisinal knowledge work. In the coming AI era my expectation is performance attribution will be far more granular than project assignment.

Opportunity Cost

There is one more prerequisite topic. For most companies inference is a shared commodity unlike factory equipment. The gpus are never idle as long as work is happening somewhere in the world. There is no capacity utilization benefit to run the business more hours of the day.

There exists however the opportunity cost of missed capabilities, growth, and learning. While utilization is incremental margin for a factory the risk of slower improvement for products and services can be existential. The startups in SF working 6 days in office seem to agree the winner of their category will be running as continuously as possible.

Relay Week

The goal of the relay week is to solve for competing interests in the AI era. The company benefits from continuous productivity. The employees need sleep and generally want more paid time off. The status quo today is projects make forward progress 40 to 60 hours per week and employees get weekends and evenings.

The core idea of the relay week is to hand off projects for 24/7 progress. The exact schedules and shifts can be arranged in multiple ways.

The first opportunity is an overnight relay. The company can set up an office in a complimentary time zone. The employees can work a regular week and projects make forward progress for 80 to 120 hours. There are operational challenges to collaborating across the world but given the need for sleep there is no better way to get continuous progress.

The second opportunity is a midweek relay. If there are two locations and each location operates 12 hours per day and through the weekend the result is uninterrupted progress. The problem is very few people are capable of working that hard and even fewer are inclined to do so. The solution can be another handoff. If projects make progress across teams overnight there's no reason projects can't make progress across teams midweek.

In a fully scaled relay there are 4 teams, 2 per location. For example teams A and B in Toronto, teams C and D in Singapore. The schedule would be: A to C back and forth for 3 days on 12-hour shifts, then midweek A to B and C to D for 6-hour shifts, then B to D back and forth for 3 days on 12-hour shifts, then B to A and D to C to start the next week.

The schedule for employees is three long 12-hour days, one 6-hour day, and a 3-day weekend. In a hard-working industry with unintended 10-hour or 12-hour days in place of 8-hour days, a 3-day weekend might be a meaningful improvement.

For employees in different stages of life it could work to reduce to 10-hour days and allow projects to stall for a couple hours or set up a 3-location relay for continuous 8-hour shifts. Another variation is alternating 3-day and 4-day weeks to avoid a half day. I assume long shifts are more productive for supervising but that is not based on actual data.

The last time we changed the work schedule was the industrial revolution. As the AI revolution gets underway it feels like a reasonable time to pause and reflect if we might enjoy a long weekend every week.