Convergence of Software Engineering Productivity in Changing Organisations
Module 2 of 6 Intermediate 35 min

Reading the J-Curve — Disruption, Recovery, and What the Endpoint Really Means

Prerequisites: Why Engineering Productivity Always Finds Its Level

What you'll learn

  • The three phases of the J-curve — disruption trough, recovery slope, and convergence endpoint — and what is happening organisationally in each.
  • Why the depth and duration of the J-curve vary across organisations and change types.
  • How to distinguish a normal J-curve dip from a structural stall, so you know when to hold course and when to intervene.
  • Why outperformance above the prior productivity baseline is possible but conditional — it only occurs when structural change accompanies the disruption.

Why this matters

You have probably lived through at least one change initiative that felt, for a painful stretch of months, like a mistake. Velocity dropped, standup meetings got longer, people were uncertain, and someone senior started asking uncomfortable questions. That moment — when the dip feels permanent — is when most organisations make their worst decisions: reversing course, switching approaches, or losing confidence in the people leading the change.

If you can read the J-curve fluently, you can do something more useful than endure that moment. You can name it, explain it to stakeholders, forecast how long it will last, and — critically — recognise the difference between a healthy dip and a genuine signal that something structural has gone wrong. That skill, more than any particular methodology, determines whether your organisation extracts value from change or repeatedly pays the cost without collecting the reward.

Core concept

In Module 01, you saw that productivity seeks its structural ceiling — the level set by team structure, cognitive load distribution, and communication patterns — and that temporary disruptions don't permanently shift that level. Here we examine the shape of the journey when disruption occurs.

The J-curve is the specific shape of the convergence curve during a single disruption event: productivity dips below the pre-change level, then recovers. The name comes from the letter itself — imagine plotting productivity on the Y-axis and time on the X-axis. The line drops, curves, and climbs. The path traces a J.

Phase 1 — The disruption trough

The trough begins the moment the change lands and ends when productivity stops declining. During this phase, unlearning dominates. This is the part that surprises people: not the challenge of acquiring new skills, but the mandatory demolition of existing ones.

Muscle memory, accumulated over months or years, actively fights the new approach. A developer who has spent two years navigating a monolith knows exactly where to look when something breaks. That tacit knowledge is now worthless. The same developer, facing a distributed microservices system, must rebuild debugging instincts from scratch — while simultaneously learning new deployment pipelines, new service communication patterns, and new conventions for ownership. None of those learning curves run in sequence. They compound.

Research confirms that the trough is deeper and longer in established organisations than in younger, smaller ones. A startup with six months of accumulated practice has very little to unlearn. An enterprise with 15 years of accumulated process has far more — and its structural inertia means the unlearning is slower and more contested. Studies of firms adopting industrial AI found some experiencing short-term losses as severe as 60 percentage points; the average dip ran to several years before recovery appeared, with smaller and newer firms recovering faster across the board.

The implication for leaders: do not measure success during the trough. It is the wrong instrument at the wrong moment.

Phase 2 — The recovery slope

The recovery slope begins when new practices start to solidify and productivity returns toward its prior level. This phase is characterised by increasing competence in the new approach combined with the gradual reconstruction of tacit knowledge, communication norms, and coordination shortcuts.

What drives recovery speed? Several factors interact:

  • How much had to be unlearned. Legacy infrastructure, long-standing conventions, and established communication patterns all create more unlearning debt.
  • The scope of what changed simultaneously. Technical change alone is one learning curve. Add procedural changes, cultural shifts, and emotional adjustment — what happens to team identity and trust — and you compound the cognitive demand. Teams navigating all four simultaneously have longer recovery slopes.
  • How clearly the new structure is defined. Ambiguous ownership, unclear reporting lines, or disputed responsibilities extend the slope because teams spend energy on coordination overhead rather than productive work.

The recovery slope is the phase where holding course matters most. Research from the Nave analysis of transformation initiatives found that leadership tolerance for the dip is the single most reliable predictor of whether the initiative ultimately succeeds. Organisations that reverse course during the recovery slope pay the full cost of the disruption trough and collect none of the benefit.

Phase 3 — The convergence endpoint

The convergence endpoint is where the curve settles: the new productivity baseline after the disruption has worked through. And here is the most important thing this module will tell you.

The endpoint can land in three places: at the prior baseline, below it, or above it. Which one occurs is not random, and it is not primarily about effort, training, or leadership messaging. The endpoint depends on whether structural change accompanies the disruption.

This is the conditional rule that the research supports — and that is frequently misread in practice. The popular version of the J-curve story says: "hold through the dip, and you'll come out above where you started." That version is too simple. Brynjolfsson, Rock, and Syverson's foundational work on intangibles and general purpose technologies shows that the recovery to above-baseline requires complementary investment — in skills, process redesign, and organisational structure — not just time and persistence.

Stated precisely: outperformance above the prior productivity baseline occurs only when structural change accompanies the disruption. Without it, the convergence endpoint lands at or below the prior baseline, because the structural ceiling hasn't moved.

Concrete example

Apex Engineering — a 200-person software organisation inside a financial services company — gives us a clean illustration. In Year 1, Apex migrated from a monolithic payments platform to a microservices architecture.

Here is what the J-curve looked like, described as a simple shape:

[Diagram: J-curve showing productivity on Y-axis (as % of pre-migration baseline), time in months on X-axis, with a trough around months 3–4 at roughly 75% of baseline, then recovery along a rising slope through months 5–7, converging near the prior baseline by month 8.]

Months 1–4 — the disruption trough. Developers learned new deployment pipelines, debugging became distributed and unfamiliar, and communication overhead spiked as teams coordinated across service boundaries. Velocity dropped roughly 25% from its pre-migration baseline. New incident patterns appeared that nobody had mental models for. The monitoring tooling was unfamiliar, and on-call rotations became more demanding.

Months 5–8 — the recovery slope. New practices began to solidify. Developers rebuilt debugging instincts for the distributed context. Teams settled into service ownership conventions. Velocity recovered toward its prior level.

Month 8+ — the convergence endpoint. Velocity converged close to the pre-migration baseline — but not above it.

Now consider what would have happened if Apex had simultaneously redrawn team boundaries to match the new service architecture — assigning each stream-aligned team clear end-to-end ownership of specific services rather than the loosely-defined system slices they inherited from the monolith. The recovery slope would likely have been similar in shape but the convergence endpoint would have landed above the prior baseline. Reduced coordination overhead, clearer ownership, and better-matched cognitive load distribution would have raised the structural ceiling. Without that structural change, the productivity baseline simply returned to where it was before — because the ceiling that governed it hadn't moved.

This contrast is not a critique of Apex's migration. The technical change was sound. The lesson is narrower: the J-curve's endpoint is determined not by whether the change was worthwhile, but by whether structural change accompanied it.

Analogy

A tennis player switching from an Eastern grip to a Western grip experiences something structurally identical to what Apex experienced. For weeks — sometimes months — consistency drops, speed decreases, and matches are lost that should have been won. The old grip's muscle memory actively interferes with the new one. The player is living in the trough.

Those who persist typically exceed their previous performance once the new grip is internalised. The mechanics are better; the swing generates more topspin; the player's game becomes more durable under physical fatigue.

But there is an important caveat before you carry this analogy into an organisational context. The individual athlete's recovery is almost always above the prior baseline — because the athlete is the structure. When she changes her grip, she changes the whole system. For an organisation, that equivalence does not hold. The individuals can learn the new approach completely, and the organisation can still converge to the prior baseline or below it — because the structural elements (team boundaries, ownership, communication patterns) haven't changed. Structural change in an organisation requires deliberate intervention beyond individual skill acquisition. We will explore what those interventions look like in Module 03.

Going deeper

The J-curve concept traces back to macroeconomics, where it describes exchange rate dynamics: when a currency depreciates, the trade balance initially worsens before improving, because contracts and prices adjust slowly. Brynjolfsson, Rock, and Syverson adapted the concept for technology adoption, coining the "Productivity J-Curve" in their 2021 American Economic Journal paper — where they showed that intangible investments (training, process redesign, organisational adaptation) are the mechanism that enables recovery and eventual outperformance. This is a direct empirical grounding for the conditional rule above: without the complementary intangible investment, the curve doesn't climb past its prior level.

The 2025 Census Bureau working paper by McElheran, Yang, Kroff, and Brynjolfsson extends this into industrial AI adoption specifically, using detailed firm-level data. Their finding is striking: firms adopting industrial AI showed causal evidence of J-curve-shaped returns, with short-term performance losses preceding longer-term gains. But the losses were selective — firms that invested in complementary adjustments (process redesign, skill development, operational restructuring) recovered and outperformed; those that treated AI as plug-and-play did not.

Distinguishing a J-curve dip from a structural stall. One practical skill this module is trying to build is pattern recognition: knowing when the dip is temporary (J-curve) versus when it has become a structural stall (the recovery slope has flattened and isn't climbing). Several signals distinguish them:

  • A J-curve dip decreases in severity over time — the rate of decline slows, then reverses. A structural stall shows flat or worsening trajectory months after the trough should have passed.
  • A J-curve dip is correlated with identifiable unlearning events — the deployment of the new system, the reorg's effective date, the moment old tools were switched off. A structural stall often correlates with persistent coordination problems that weren't addressed when the change landed.
  • Recovery slope progress is detectable through leading indicators — team confidence, incident resolution time, cycle time per task — before lagging indicators like overall velocity recover. A structural stall often shows no movement in leading indicators despite passage of time.

When you observe a structural stall, the right response is diagnosis, not patience. The question is which structural constraint is preventing recovery — a topic that Module 03 explores in detail through the lens of cognitive load and team topology.

Common misconceptions

"The dip means the change was wrong." This is the most dangerous misconception in practice, because it motivates the most costly response: reversing course during recovery. A J-curve dip is not evidence of failure; it is evidence of adjustment. The test is trajectory, not current level. If the dip is occurring and the rate of decline is slowing, the change is likely working. If the trajectory is still worsening months after the trough should have passed, that is the signal worth acting on. Reversing course midway typically locks in the worst outcome: the disruption cost is paid, the benefit is forfeited, and a second disruption begins the moment the reversal is implemented.

"Holding through the dip guarantees outperformance." The J-curve story is sometimes told as an unconditional promise: persist, and you'll come out ahead. Research does not support the unconditional version. Outperformance above the prior productivity baseline is conditional on structural change accompanying the disruption. Teams that acquire new skills and rebuild their workflows in the same structural configuration as before will converge to approximately the same baseline as before — because the structural ceiling that governs the baseline hasn't moved. Persistence is necessary but not sufficient.

"Good training prevents the J-curve." Comprehensive training can reduce the severity and duration of the trough. It cannot eliminate it. The J-curve is not primarily a knowledge gap; it is a muscle memory and structural adaptation gap. Even well-trained teams must rebuild tacit knowledge, re-establish communication norms, and develop new coordination shortcuts. Training accelerates the recovery slope — which is genuinely valuable — but it does not flatten the curve to zero.

Check your understanding

  1. A team deployed a new incident management system three months ago. Velocity is down 18% from pre-deployment. A senior stakeholder is asking whether to roll back. What questions would you ask to determine whether this is a J-curve dip or a structural stall, and what would the answers tell you?
Reveal answer You would want to know the trajectory of the decline: is the 18% figure worse, better, or flat compared to weeks two and four post-deployment? A J-curve dip should be showing a slowing rate of decline or early recovery by month three. You would also look at leading indicators — cycle time per incident, mean time to resolution, team confidence in the new tool — which should show some improvement before overall velocity recovers. If the trajectory is still declining at month three with no improvement in leading indicators, that is a structural stall signal worth investigating. You would also ask what structural changes accompanied the deployment: were ownership and escalation paths clearly redefined, or did the new tool land on top of the old process? The answer to that question predicts whether the convergence endpoint will sit at or above the prior baseline.
  1. Apex Engineering's microservices migration converged back to its prior baseline rather than above it. What structural change would have been required for the endpoint to exceed the prior baseline, and why does that variable matter?
Reveal answer Apex needed to redraw team boundaries to match the new service architecture — assigning each team clear, end-to-end ownership of specific services rather than loosely-defined slices inherited from the monolith. This matters because the **structural ceiling** — the maximum sustainable productivity level given the organisation's team structure, cognitive load distribution, and communication patterns — hadn't changed. The technical migration was sound, but the underlying structure that governs the **productivity baseline** remained the same. Without raising the structural ceiling, convergence returns to the old baseline by definition.
  1. Why is the J-curve deeper and longer in established organisations than in newer, smaller ones? What does this imply for how leaders should calibrate expectations when planning transformation in a mature organisation?
Reveal answer Established organisations have more accumulated practice to unlearn — more deeply embedded processes, legacy infrastructure, and long-standing communication conventions. Each of those represents unlearning debt that must be paid before new approaches can take hold. Smaller, newer organisations have less of this debt. The implication for leaders is that recovery timelines should be calibrated to organisational maturity, not to the inherent complexity of the change alone. A 200-person organisation with 10 years of accumulated process should expect a materially longer trough and recovery slope than a 30-person team with two years of history adopting the same change. Telling stakeholders the recovery will take "a few weeks" in the first case is setting up a crisis of confidence that arrives predictably.
  1. The sports analogy — a tennis player changing grip — is frequently used to explain the J-curve. Where does the analogy hold, and where does it break down when applied to organisations?
Reveal answer The analogy holds well for the trough: muscle memory actively interferes with new approaches, performance dips before it recovers, and persistence is required to reach the recovery slope. Where it breaks down is at the convergence endpoint. An individual athlete who changes technique reliably ends up above her prior baseline if the new technique is genuinely better — because *she* is the whole system, and changing her approach changes the entire system. In an organisation, individuals can fully internalise the new approach and the organisation can still converge to the prior baseline, because the *structural* configuration — team boundaries, ownership, communication patterns — is a separate variable that requires deliberate intervention. Organisational outperformance requires structural change; individual outperformance does not.
  1. Why is "the leadership's tolerance must exceed the pain of change" considered a critical predictor of J-curve outcomes? What does leadership tolerance look like in practice during a trough?
Reveal answer During the trough, stakeholder pressure to reverse course is at its highest — precisely when reversal is most costly. Leaders who abandon the initiative during the trough pay the full cost of the disruption without collecting any benefit, and often trigger a second disruption when they implement the reversal. In practice, leadership tolerance means: communicating a clear rationale for why the dip is expected and temporary; providing stakeholders with trajectory data (not just current-state snapshots); preserving investment in the initiative rather than cutting resources at the moment of maximum apparent cost; and distinguishing publicly between J-curve dip (expected) and structural stall (actionable concern). Leaders who confuse the two, or who communicate uncertainty about the initiative during the trough, accelerate the loss of stakeholder confidence that causes early abandonment.

Key takeaways

  • The J-curve has three phases: disruption trough (unlearning dominates), recovery slope (new practices solidify), and convergence endpoint (the new productivity baseline). The trough is not a sign of failure — it is a structural feature of adjustment.
  • Depth and duration of the J-curve vary with organisational maturity and change scope. Established organisations with legacy infrastructure and accumulated practice have more to unlearn, and their curves run deeper and longer.
  • The convergence endpoint — at, below, or above the prior baseline — is conditional on structural change. Outperformance requires that the structural ceiling be raised, not just that teams persist through the trough.
  • Distinguishing a J-curve dip from a structural stall requires trajectory analysis and leading indicators, not snapshots of current productivity. The right response to a stall is diagnosis; the right response to a dip is patience and stakeholder communication.
  • The sports analogy (grip change) illustrates the trough well but breaks down at the endpoint: individual skill change reliably yields outperformance; organisational change only does so when accompanied by deliberate structural redesign.

References