Work-in-Progress Limits in DevOps: Where the Practice Stands Today

Rajesh Kumar

Rajesh Kumar is a leading expert in DevOps, SRE, DevSecOps, and MLOps, providing comprehensive services through his platform, www.rajeshkumar.xyz. With a proven track record in consulting, training, freelancing, and enterprise support, he empowers organizations to adopt modern operational practices and achieve scalable, secure, and efficient IT infrastructures. Rajesh is renowned for his ability to deliver tailored solutions and hands-on expertise across these critical domains.

Few ideas in software delivery have travelled as far as the humble limit on work in progress. It began on factory floors, moved into lean thinking, and now appears on team boards from small startups to large enterprises. Here is a survey of where the idea came from, what the available evidence supports, where adoption stalls, and what Japanese and international DevOps teams can reasonably expect from it today. It avoids headline numbers, because the honest picture is qualitative and varies by team.

Where the idea came from

The practice has a clear lineage. In postwar Japanese manufacturing, the Toyota production system introduced kanban cards as a signal: a station produced more only when a card showed that downstream demand existed. The cards limited how much unfinished material could pile up between stages. That was a work-in-progress limit in physical form, and it was a deliberate response to the cost of holding half-made goods.

Decades later, practitioners of lean and agile software methods adapted the idea. Boards with columns replaced cards on trays, and a number written above a column replaced the rule that a station could not produce without a signal. The principle stayed the same: pull new work only when there is room, and prefer finishing over starting.

What the evidence actually supports

There is a well-established mathematical relationship between the average number of items in a system, the rate at which they are completed and the average time each spends there. This result, usually called Little’s Law, is a property of queues in general. Applied to a team board, it says that if completion rate stays roughly constant, then more items in progress means longer average time to finish each one. That is not an opinion about teams; it is arithmetic about queues.

What the law does not tell you is whether a given cap will improve a given team. Completion rate is rarely constant, and people are not machines. Studies and practitioner reports commonly describe benefits such as shorter lead times, earlier feedback and fewer forgotten items, but results depend on context, and anecdotes should not be mistaken for universal proof. A sensible reading is that the mechanism is sound and the size of the effect has to be measured locally.

The current state: wide awareness, uneven practice

Most DevOps teams know of the concept. Fewer apply it strictly. The typical pattern is a board with a limit that exists on paper but is ignored when requests arrive, because saying “not yet” is socially harder than adding another card. In many organisations the limit is also set once and never reviewed, so it either drifts upward until it means nothing or stays so tight that people work around it.

Another common pattern is counting only visible engineering tasks. Operational requests, meetings with stakeholders, informal favours and on-call interruptions all consume attention but rarely appear on the board. A limit that ignores them understates the real load. Teams that measure honestly tend to find their working capacity is lower than the board suggests.

Why adoption stalls

Idle time feels wasteful

A person who cannot start new work because the limit is reached may seem idle. The cultural answer is to help finish something already open, perhaps by reviewing, testing or unblocking. That requires a collaborative team culture, which is why the practice often travels together with broader change work, as discussed in the piece on how to manage cultural change when adopting DevOps.

Demand is presented as urgent

When every request is labelled urgent, a limit looks like obstruction. The remedy is a visible queue with an agreed order, so that urgency is handled by changing the order, not by adding to the load.

Measurement is thin

Without even a simple record of start and finish dates, nobody can show whether the limit helped. A team that writes both dates for each item for a few weeks will have enough to judge the effect on cycle time.

Looking ahead for DevOps teams

The direction of travel for DevOps practice, in Japan and elsewhere, points toward more automation, more platform tooling and more observability. These advances shorten certain steps, but they do not remove the problem of too many open items; in some ways they make it worse, because starting a new automated task is cheaper than ever. The broader outlook is explored in the article on the outlook for DevOps across Japan and the wider world. If starting gets easier while finishing stays hard, then deliberately capping what is open becomes more valuable, not less.

Beyond engineering

It is worth noticing that the same habit shows up outside technology. Personal productivity discussions, household planning threads and wellness communities all circle around the question of how many things one person can keep going at once. The wellness-themed website AskDoctorLive runs member discussions and articles next to listings for hospitals and doctors, and is one of countless places online where everyday routines get raised. That breadth is a reminder that the principle is human, not merely technical.

A measured way to try it

  1. Record start and finish dates for every item for three weeks without changing anything.
  2. Set a limit near the number of people on the team and display it.
  3. Define the columns clearly and count items until they are truly finished.
  4. Record the same dates for another three weeks.
  5. Compare average time to finish and the number of items completed.
  6. Adjust the limit according to what the data shows, and write down why.

Reading the results with care

Do not expect a dramatic change in every team. Some will see shorter waits straight away; others will learn first that their real constraint is a shared review step or a dependency on another group. Both outcomes are useful. A limit makes the constraint visible, and visibility is what allows a team to make a rational decision about it.

The state of the practice, then, is mature in theory and uneven in daily use. The idea is old, the mathematics is sound, and the human difficulty is as real as ever. Teams that combine a modest limit with honest measurement and a willingness to say “not yet” are the ones most likely to see the benefit, and they can start with nothing more than a board and a number.

Leave a Reply