Agile Constraints Management
Pattern: A named solution to a recurring problem.
Steve Tendon’s integration of Agile software development with the Theory of Constraints: cut the backlog into economically assessable target scopes, forecast each one from delivery history, protect the forecast with a critical-chain buffer, and release work only when the constraint can take it.
Also known as: TameFlow; Constraints Management and Throughput Economics for Agile Software Development.
A large software organization can run forty Agile teams that all report healthy velocity, short cycle times, and steady delivery, while the executives paying for them still can’t say what the investment did to revenue, cost, or return. Agile Constraints Management is Steve Tendon’s answer to that gap. It keeps iterative, incremental delivery, adds the constraint and the Throughput Economics decision frame, and replaces expert estimates with forecasts drawn from the system’s own delivery history. Tendon has published the approach under the TameFlow name since 2014. TOCICO recognized his 2025 whitepaper on it as part of the TOC Body of Knowledge.
Context
Agile methods solved a real problem: software scope is discovered while the work is done, and small teams delivering working increments learn faster than teams executing a fixed plan. The trouble starts at scale. A large organization has many teams, shared specialists, approval steps, and test environments, and the constraint often sits between teams rather than inside one. Tendon argues that Agile copes with local constraints but not with systemic, hidden ones, and that its estimates hold only where effort is most of the flow time. In complex projects, he argues, waiting dominates.
TOC already covers parts of the problem in software: ToC for Software and DevOps for the delivery pipeline, Flow Metrics for Knowledge Work for measurement, Throughput Economics for Software Organizations for roadmap decisions, and Agile CCPM for release commitments. Agile Constraints Management combines flow measurement, CCPM-style commitment, and Throughput Economics in one operating method that reaches down to the backlog.
Tendon derived it with Goldratt’s Standing on the Shoulders of Giants analysis: take a widely adopted body of practice, name the area it leaves unaddressed, expose the assumptions that block it, and inject TOC where those assumptions fail. Eli Schragenheim and Graham Scott advised on the paper.
Problem
Agile measures operating output, not economic result. Tendon separates the two senses of one word. Economic Throughput ($T) is TOC’s Throughput: the rate at which the business generates money through sales. Operational Throughput (TP) is the Agile and Kanban sense: the rate of delivery of customer-valued work items. A team can double its TP and leave $T where it was. Velocity, burndown, and story counts can’t tell an executive where the next quarter of engineering should go.
Three questions stay unanswered. Where is the constraint across all these teams? When will a given piece of work actually be done, waiting included? And which work should go through the constraint first, judged by its effect on throughput, investment, and operating expense?
Forces
• A backlog never ends. Without a line drawn across it, there’s nothing finite to forecast and no unit whose economic effect can be assessed.
• Expert estimates miss the waiting. Flow time is effort plus wait. Teams estimate effort, which in Tendon’s analysis of complex projects is the smaller part of the elapsed time.
• Timeboxes do useful work. Sprints give teams a rhythm, a review point, and an implicit limit on work in process. Removing them takes something away that has to be replaced.
• Not all valuable work earns money directly. Architecture, research, and defect repair protect or enable future throughput without generating any this quarter.
• The constraint is intangible and can move. It may be a scarce specialist team, a slow approval, or a shared test environment, and it shifts as the portfolio changes.
Solution
Partition the backlog into MOVEs, forecast each MOVE from flow history, manage the forecast as a critical chain with a time buffer, release MOVEs only when the constraint has capacity, and choose among them by their effect on throughput, investment, and operating expense.
The paper names two main injections and one supporting injection.
The MOVE. A MOVE is a target scope: a bounded subset of the backlog that delivers a defined business benefit, either money or an observable outcome. The paper’s definition note (p. 42) expands it as “Minimal Outcome-Value Effort”; its injection list (p. 19) writes “Minimum Outcome Value Effort.” The benefit belongs to the organization building the software, not the customer, and one MOVE may take several user stories. The paper sorts MOVEs by what they do to the TOC measures:
| MOVE type | Changes $T directly? | What it is for |
|---|---|---|
| Value | Yes | Generates throughput; ranked by $T per unit of constraint time |
| Outcome | No | Produces an observable effect that isn’t measured in money |
| Development precursor | No | Must be built before another MOVE can start |
| Delivery precursor | No | Must ship first for commercial or regulatory reasons |
| Architectural element | No | Supports the architecture; its OE and I may carry over to the first MOVE it enables |
| Exploratory | No | Buys knowledge for a go/no-go decision, capped by a money box rather than a timebox |
| Failure demand | Yes | Repairs defects or technical debt; valued as the throughput lost if it is left undone |
Mutually exclusive alternatives are modeled as “XOR MOVEs,” so giving up an option is costed too. Green MOVEs sustain throughput today, and red MOVEs protect or build future throughput, which Tendon ties to Goldratt’s stability and growth curves.
Flow metrics and probabilistic forecasting. Because a MOVE is finite, its completion can be forecast from the delivery history of the teams that will do it. The paper credits the technique to Daniel Vacanti’s ProKanban work. Engineers stop supplying expert estimates. The forecast timeline is treated as the equivalent of a critical chain and protected by a time buffer, so Fever Charts and the buffer burn rate (percentage of buffer consumed over percentage of work done, Tendon’s wording of a measure the TOCICO Dictionary defines against critical chain completed) apply to each MOVE, each team’s share of a MOVE, or many MOVEs at once, as in multi-project CCPM.
S-DBR or DBR release. The supporting injection limits work in process. New MOVEs enter the workflow only when a drum signal shows capacity at the constraint. The paper names both Simplified Drum-Buffer-Rope and full Drum-Buffer-Rope for this without choosing between them. Fixed timeboxes and cadences are dropped. Teams converge when a “Work Execution Signal” calls for it, and the CCPM buffer signal is the main such signal.
Selection then follows Throughput Economics. Value MOVEs are ranked by $T per unit of constraint time; the other types are judged by their effect on operating expense and investment.
Where the method is contested
Estimation. Two TOCICO-recognized bridges from CCPM to software disagree. Ujigawa and Updegrove’s Agile CCPM keeps relative size estimates, velocity, and a scope buffer. Tendon drops estimation entirely and aligns the method with the #NoEstimates movement, though “backed with much better science.” Tendon’s paper lists the Agile CCPM paper in its references but doesn’t discuss its estimation approach or report a comparison.
Little’s Law. Tendon writes that probabilistic forecasting “is entirely based on Little’s Law” and argues that the method is scaleless because the law holds whatever the flow-time distribution, number of servers, or order of service. Little’s 2011 account supports the generality: the relation among average work in process, arrival rate, and time in system holds exactly over an observed period and doesn’t require stationary conditions. But Little also says the law puts its user “in the measurement business, not the forecasting business”; forecasts come from statistical methods applied to the data. The law keeps the flow measures consistent. The forecast rests on future delivery resembling past delivery, which limiting work in process makes more plausible. Scalelessness is Tendon’s claim about the method, not an established property of forecasts.
Timeboxes. The paper calls timeboxes and cadences “artificial and arbitrary.” Scrum and SAFe practitioners will reply that sprints are how many teams get stability, and the paper itself concedes that timeboxing is how Agile usually limits work in process. Tendon substitutes the drum for the sprint. Agile CCPM also rejects fixed timeboxes, on Parkinson’s Law grounds.
Inherent simplicity. Tendon argues that properly engineered software, built from components with clean interfaces, offers “multiple degrees of freedom,” while Goldratt’s Inherent Simplicity presupposes systems with few. On that reading the governing constraint is harder to find, and the paper doesn’t settle how to find it.
Evidence. The paper reports no implementation results, and searches through October 2026 found no independent or peer-reviewed evaluation of the method. The available sources are Tendon’s paper, talks, and books, plus interviews with him.
TOCICO recognition is not outcome evidence: TOCICO’s recognition places the paper in the TOC Body of Knowledge. It’s a judgment about fit with the field’s concepts, not a finding that the method improves delivery or profit. Treat any reported gain as a claim to test against your own flow and financial measures.
How It Plays Out
The paper contains no worked case, so the two scenarios below are illustrative.
A payments software company runs 14 product teams. Its flow metrics show work waiting longest in front of a five-person security-review group, the likely constraint. Product leadership cuts the merchant-onboarding backlog into MOVEs. Self-serve onboarding for small merchants is a Value MOVE with an expected $T. A new identity-verification service is a development precursor. A fraud-scoring prototype is an exploratory MOVE capped at a fixed spend. A recurring settlement defect is a failure-demand MOVE, valued by the throughput it costs each month. Value MOVEs are ranked by expected $T per hour of security review, and a MOVE enters the workflow only when security review can take it. Five weeks into onboarding, the fever chart shows 60 percent of the buffer consumed against 35 percent of work done, a burn rate of about 1.7. The teams on that MOVE stop new work and swarm the blocked items at security review instead of waiting for a sprint boundary.
A year later, a reorganization merges two teams and doubles their work in process. Their forecasts start missing by weeks, because the history they’re drawn from no longer describes the system. Better estimation wouldn’t help. The leadership restores the release limit and waits for the merged team to build a stable history.
Consequences
Benefits:
• Agile delivery connects to money. Each MOVE carries an explicit effect on $T, OE, and I, so the portfolio can be argued in Throughput Economics terms rather than story counts.
• Engineers stop estimating. Forecasts come from observed flow, including the waiting that effort estimates miss.
• One control signal works at every level. Buffer burn rate is a ratio, so a single team’s MOVE and a program of nested MOVEs read on the same fever chart.
• Work in process is held to the constraint’s pace. Drum-based release keeps teams from flooding the shared specialist group.
• Non-revenue work gets an economic treatment. Architecture, research, and defect repair are costed as precursors, exploration, or avoided loss instead of competing on persuasion.
Liabilities:
• The economics are only as good as the $T estimates. A Value MOVE needs a throughput forecast, which can become a guess relabeled as an economic case.
• Forecasts need a representative history. New teams, new kinds of work, and surges in work in process break the link between past and future flow.
• Dropping timeboxes removes a familiar rhythm. Review, demonstration, and learning need another trigger, and the method relies on signal discipline to provide it.
• The vocabulary comes from one source. MOVE types, XOR MOVEs, and work execution signals are Tendon’s terms, developed through his consultancy, and “TameFlow” is a registered trademark.
• Finding the constraint stays hard. The method names the difficulty in loosely coupled systems, but flow metrics alone don’t settle where the governing constraint is.
Related Articles
Complements: ToC for Software and DevOps — ToC for Software and DevOps applies TOC to the delivery pipeline; Agile Constraints Management adds the work-packaging unit and the economic selection rule for what enters that pipeline.
Contrasts with: Agile CCPM — Both are TOCICO-recognized bridges from CCPM to software, but Agile CCPM keeps relative size estimates and velocity while Agile Constraints Management drops estimation for probabilistic forecasting.
Depends on: Flow Metrics for Knowledge Work — Flow time, work in process, and delivery throughput are the history the method forecasts from, and its split between economic and operational throughput matches the flow-metrics distinction.
Depends on: Throughput — The method's economic throughput is TOC Throughput, money generated through sales, kept distinct from the count of delivered work items.
Extends: Throughput Economics for Software Organizations — Agile Constraints Management carries the T, I, and OE decision frame from roadmap and portfolio moves down to backlog scope through the MOVE.
Implements: Five Focusing Steps — The method adapts the focusing steps to intangible software constraints such as a scarce specialist team, a slow approval, or a shared test environment.
Related: Inherent Simplicity — Tendon argues that well-architected software has so many degrees of freedom that Goldratt's search for one governing constraint becomes harder.
Related: Microsoft XIT Drum-Buffer-Rope Implementation — Anderson's XIT case is the earlier documented use of DBR release control on software work, the same mechanism this method uses to release MOVEs.
Uses: Critical Chain — The forecast timeline of a MOVE is treated as the equivalent of a critical chain, so CCPM buffer management applies to it.
Uses: Fever Chart — Fever charts track each MOVE, or each team's portion of a MOVE, by buffer consumed against work done.
Uses: Project Buffer — A time buffer protects the forecast completion of each MOVE in the way a project buffer protects a critical chain.
Uses: Simplified Drum-Buffer-Rope (S-DBR) — S-DBR or DBR is the supporting injection: MOVEs are released only when the drum signals capacity at the constraint, which limits work in process.
Uses: Throughput Economics — Value MOVEs are ranked by throughput per unit of constraint time, and the other MOVE types are judged by their effect on operating expense and investment.
Sources
• Steve Tendon (advisors Eli Schragenheim and Graham Scott), “Constraints Management and Throughput Economics for Agile Software Development” (TOC Body of Knowledge whitepaper, Revision 1, 9 August 2025) — the TOCICO-recognized primary source for the invalid assumptions, the MOVE and its types, probabilistic forecasting with buffer management, the S-DBR/DBR release rule, and the scalelessness argument.
• Steve Tendon and Wolfram Müller, Hyper-Productive Knowledge Work Performance: The TameFlow Approach and Its Application to Scrum and Kanban (J. Ross Publishing, 2014) — the book-length origin of the TameFlow approach.
• Ujigawa and Updegrove, “Agile CCPM: Critical Chain for Software Development” (TOCICO White Paper Series, 2016) — the TOCICO-recognized alternative that keeps relative size estimation, velocity, and a scope buffer.
• John D. C. Little, “Little’s Law as Viewed on Its 50th Anniversary”, Operations Research 59(3), 2011 — the law’s author on what L = λW states, its conditions, and why it is a measurement relation rather than a forecasting method.
• Schragenheim, Camp, and Surace, Throughput Economics (2019) — the decision frame the method applies to MOVE selection.
• Goldratt, “Standing on the Shoulders of Giants” (Gestão & Produção 16(3), 2009) — the analysis process Tendon follows to derive the method from Agile’s invalid assumptions.
• Cox, TOCICO Dictionary (2nd ed., 2012), entries “buffer burn rate,” “fever chart,” “throughput,” “throughput per constraint unit (T/CU),” and “simplified drum-buffer-rope (S-DBR)” — the certification-grade vocabulary the method reuses.