I’d like to preface this letter with appreciation for Comrade Firmiani’s thoughtful engagement with Dónal Ó Coisdealbha’s essay for Cosmonaut Magazine on the Toyota Production System and cybernetic planning. The original essay is very provoking, and it fills in a gap in socialist discourse many, myself included, have struggled to fill. I agree with Comrade Firmiani about the limits of this particular essay however, that it does not adequately address the conflict between local and global optimization and remains too focused on the elements of management science within the point of production and so only indirectly points to the necessary work socialist planners would face in a new society.

However, I do think Comrade Firmiani makes a mistake in assuming that Comrade Coisdealbha’s work represents the limits of cybernetic or management scientific literature on the subject. This is an understandable error, as unfortunately explicitly ‘cybernetic’ management science literature has been rather sparse since the 1990s. Not that it doesn’t exist, journals like Kybernetes among others still publish regular pieces, but compared to other sub-disciplines, like industrial psychology, management cybernetics has been quite unfashionable, since the larger Cybernetic mothership was dismembered for parts in the academy, with its empirical insights incorporated into other fields, while its philosophical and meta theoretical interventions were left to yellow in thrift book shops. Fortunately, with the revival of interest in Cybersyn thanks to many authors like Medina and Morozov, and I’d like to think in some small part Cosmonaut Magazine, there’s some hope that it may go beyond intellectual archeology.

Back to Comrade Firmiani’s letter, they end the piece with the passage:

These issues are inherent to local optimization, where information sharing is not centralized. A global view of the system will always be necessary, since local interventions are only capable of optimizing certain types of problems. The challenge of socialist planning will be to reconcile this centralization and decentralization in a democratic fashion.

I have good news! This is precisely one of the key insights within Stafford Beer’s management cybernetics (as well as one of the central concerns of the Soviet economic cybernetics of Kitov and Glushkov). While the Viable System Model does propose that maximal effective local autonomy is necessary for a resilient organization, it also proposes highly integrated information monitoring, auditing and regulation. I hesitate to use the term centralized here because Beer is careful to emphasize that central planners in the real world necessarily face a cognitive resource asymmetry vs the scope of the real system they're trying to manage, and so normally should be mainly dealing with high-level indicators; but in principle even granular local performance data should be accessible to central planners should a threat escalate to their level. Beer would, on a personal level, likely caution that this information should be handled in such a way as to protect the privacy and freedoms of shop floor workers.

Moreover, because each level of management in a given economic ‘recursion’ (Beer envisioned the invariant elements of the essential model of an organization as a recursive fractal) should have real-time monitoring capabilities, their primary planning function would be in using computer models to forecast short, medium, and long-term futures the subordinate elements may face. This is quite different from either Soviet efforts to use linear programing for planning ex ante to prescriptively determine optimal resource distribution, or the bourgeois post factum use of price signals to darwinistically condition corporate planners. A planner as envisioned in management cybernetics, despite likely having some engineering training, is more like a military commander than an engineer in practice. Both are constantly ‘playing chicken with a foggy cliff’ and must have contingencies in place to rapidly intervene. For instance, during a kaizen effort that reduces inventories available to producers at a given set of work sites, plant management might have surplus inventories available in reserve to disperse, the way a Napoleonic officer might keep a company of riflemen back until needed while sending their cavalry vanguard ahead to attack the enemy position. We can wear a harness and rope as we explore foggy seaside ridges. Notably, the cybernetic ‘Ops Room,’ famously visualized in pictures from Chile’s aborted socialist experiment, was explicitly modeled on the War Rooms of Allied militaries during the Second World War.

It’s true that it is much more difficult to hold reserves with large equipment in a production setting than with inventories or labor, especially given the fact that an explicit aim of Toyota Production System (TPS)/Lean is to reduce necessary constant capital outlays (such as on equipment), however this kind of buffer-management remains possible at the next level of recursion up from the factory. In a socialist cybernetically managed economy, planners may keep a buffer of ‘redundant’ factories which run with the minimum necessary level of operation and can be scaled, using reserve labor, to address the shortages (while avoiding falling into errors like the bullwhip effect). Rather than this reserve being an imposition on workers by forcing them to involuntarily move where needed, it is more likely that planners/the democratic community would have foreseen these risks and hired people specifically for this reserve role well in advance, much like how modern states hire part-time reserve soldiers in case of a military crisis. The reason capitalist firms don’t do this is because they are driven by profit and chiefly concerned with short-term cost cutting and increases in profits, while largely being insulated from the human costs of disruptions. Under a socialist system, the principles that underlie TPS and other systems influential to modern cybernetic socialists would not be manifested in the same way they are under our current economic system.

While I am uncertain if Stafford Beer or other writers within management cybernetics directly laid out technical approaches to the problems of buffer management and how to reconcile Lean manufacturing’s pursuit of local optima with the health of the wider system, as I’m only 4 Pints of Beer in out of 10, one thinker who does is the physicist-turned-management-novelist Eliyahu M. Goldratt. Most famous for his business novel The Goal (A fantastic exploration of a slice of management science through narrative saddled with a very mediocre romance story), Goldratt applied physics and systems principles to manufacturing, supply chain, and project management across several novels, essays and lectures.

Goldratt offers a different systems approach to manufacturing than TPS which has many loyal adherents today, though fewer than those sworn to Lean. Instead of a monomaniacal focus on optimization at every work station in the plant, Goldratt points out that any efficiencies that are created anywhere but the bottleneck of the production process are a waste of effort and actually a source of potential instability and increased costs. Incidentally, this is the same principle articulated half a century earlier by Marxist systems thinker Alexander Bogdanov, which he called ‘The Law of the Least’ as a part of his proto-cybernetic Tektology. At the time, Goldratt was chiefly focused on criticizing the management common sense that all productive capacity at every work station needed to be under maximum resource exploitation in the name of ‘efficiency’, despite the fact that this created enormous surpluses of work-in-process (WIP) among other problems. Instead, Goldratt proposed what he called the ‘drum-buffer-rope’ approach, where the bottleneck set the pace of production like a drum, a sufficient buffer of work-in-process was maintained at all times so the bottleneck was never idle, and signals were sent ‘upstream’ in the process pulling work-in-process toward the bottleneck as it was consumed. Importantly, while Goldratt emphasizes working to improve the production bottleneck (which may cause it to shift elsewhere in the factory) to increase overall system efficiency, once the factory’s capacity sufficiently exceeds demand, the bottleneck transforms into a vital piece of planning infrastructure and is protected until the factory’s environment changes sufficiently.

Lean and TPS seek to solve similar problems like WIP, but they attempt to improve efficiency across the factory all at once, all the time, which leads to rapidly shifting bottlenecks and often constant instability. This is the source of the ‘whack-a-mole’ problem Comrade Firmiani very rightly identifies. Goldratt’s ‘Theory of Constraints’ (ToC) approach described above, avoids these problems by not attempting to remove too much surplus capacity in non-bottleneck processes because it seeks to avoid disruption through statistical and exceptional fluctuations. Goldratt cautioned early Lean practitioners against over-utilizing workers at seemingly idle work stations, by assigning them to other tasks, or cutting the workforce to its bare bones, because each step of the production process should have sufficient labor capacity to absorb inevitable spikes in demand to avoid cascading failure.

ToC and Drum-Buffer-Rope (DBR) aren’t without their critics, especially from partisans of Lean. They point out that DBR doesn’t offer the same kind of granularity in determining what each work site should produce that kanban does. DBR simply sends a signal that downstream work stations are accepting WIP, whereas in TPS’ Kanban, each different part is given its own card that is sent upstream when a replacement is needed. Plus, ToC traditionally has faced difficulty in implementation when the bottleneck of a production process shifts rapidly. If the wrong bottleneck is identified, or one is artificially created, there are numerous problems that can arise and cause their own sorts of chaos. However, as long as one isn’t too strictly orthodox in their commitment to Goldratt’s model, these sorts of challenges are solvable and a hybrid between Lean and ToC can emerge that takes the best of both worlds as advocated by Dee Jacobs and Suzan Bergland in their novel Velocity.

A last point on Goldratt’s relevance here is that he also addresses the issues of buffer management and the poison of local over-optimization extensively in works like his lecture series Beyond the Goal, and his rather philosophical socratic discourse The Choice. For Goldratt it is vitally important that production capacity always exceeds demand rather than match it like management common sense dictates. Moreover, Goldratt recognizes the rather cybernetic notion that Variety engineering (he uses other terminology) becomes easier the greater the scope you’re working with. He gives the example of how a customer at a car dealership may want a variation of a car not offered at a dealership. It isn’t feasible for a dealership to have every possible variation, so they only store the models most likely to meet the greatest level of consumer preference. However, if the customer is willing to wait a few days, a regional warehouse, which can maintain a wider inventory capable of absorbing even more of the demand from those not serviced by the local dealerships, can likely have the car they want. And if they still aren’t satisfied because they want a very niche and unlikely model, they could wait a few months and have a custom assembled vehicle for little to no additional cost as their specifications are sent to the assembly plant. While the idea of multi-layer buffering is commonplace today, with companies like Amazon using it (along with many other advancements in inventory management), the notion was quite revolutionary in the 1980s when Goldratt proposed it. Goldratt also prefigured Amazon in his rants against his contemporary non-integrated supply chain system which lacked system level planning. The essential point here is that the kinds of problems that a myopic application of Lean would potentially create in socialism are already addressed within management cybernetics and management science literature to varying degrees.

If we adopt a system of adequate and active buffer management, there is some danger of losing many of the real virtues TPS and Lean offer in quality improvement by removing the focusing effect that inventory and resource cuts create in Kaizen. However, this can still be deliberately created, just with failsafes available. Or we can use other methods like Netflix’s ‘chaos monkey’ approach, where they deliberately sabotage their own IT systems in unpredictable ways to force themselves to identify faults and continually improve.

The important thing here is for the socialist movement to metabolize the insights of bourgeois social science in production and transform it on a proletarian scientific basis.

I hope that Comrade Firmiani can forgive me using their letter as an opportunity to stand on a soap box, and I hope that they find my reply relevant enough to their interests and intent of their letter that it serves them some use. I certainly appreciated their intervention despite the inadequacies I felt it contained. I yearn for the day when Marxist production theory is as robust a discourse as Marxist macro economic theory or literary theory.

In Solidarity,

Amelia Davenport