TBM 416: Investment Stewardship (As Habit)
John Cutler from The Beautiful Mess · email · 2026-04-15 Reader 🖥️
Understanding the return on engineering investment is not a calculation you can perform retrospectively but a set of stewardship behaviors you must practice continuously, and if you haven't been building that habit all along, you cannot cold-start it when the question becomes urgent.
practical · argumentative — The piece argues for a specific stance (investment stewardship as ongoing habit rather than one-time calculation) and concludes with concrete action steps. It is primarily argumentative with a practical orientation: the author wants leaders to change their behavior.
Structure
| Part | Role |
|---|---|
| Opening: The question every company asks | Problem statement |
| Thesis statement (bold paragraph) | Central claim |
| You Can't Cold-Start This | Negative argument |
| Why It's Hard (And Why That's the Point) | Concession and reframing |
| What the Habit Actually Looks Like | Positive argument and characterization |
| The Job | Conclusion and call to action |
Key terms
- investment stewardship
A continuous practice of caring for and being accountable for engineering investment through behaviors like tracking leading indicators, Bayesian updating of confidence, hiring discipline, and honest finance conversations -- not a one-time ROI calculation - cold-start
Attempting to retroactively answer the engineering ROI question without having built the prerequisite habits, data, and organizational culture over time - the habit
The ongoing organizational practice of questioning leverage, tracking leading indicators, updating confidence in hypotheses, and maintaining hiring discipline -- as opposed to periodic measurement exercises - Lucas Critique
Applied here to mean that when you change organizational regime (from feature factory to fiscal stewardship), all historical relationships, incentives, and data from the old regime become invalid - good-enough proxy
A directionally correct, honestly held measurement that increases confidence in underlying hypotheses, as distinguished from both bad proxies (story points) and demands for perfect precision
Propositions
- explicitdevelopedUnderstanding the return on engineering investment is a set of continuously practiced behaviors, not a calculation you perform.conf: highextends ←TBM 430: Incubate, Compound, Refinance, Liquidate · p1 — TBM 416 argues ROI understanding is continuously practiced behavior, not a one-off calculation; TBM 430's 2x2 gives a candidate structure (quadrant position, tracked over time) for what that continuous practice could actually track.
- explicitcorroboratedYou cannot cold-start engineering ROI measurement; if you haven't been building the habit all along, you're asking too late.conf: highsupports ←Brain Food: Building Trust in Yourself Every Day · p3 — Both reject cold-starting: the capacity — ROI stewardship there, self-trust here — must be built continuously beforehand and cannot be produced at the last second.
- explicitMost companies only ask the ROI question after years of dysfunction (hiring around problems, feature factories), making the question unanswerable in the terms they want.conf: high
- explicitWhen you change organizational regime, historical data and relationships from the old regime become invalid (Lucas Critique applied to organizations).conf: high
- explicitdevelopedA bad proxy (like story points as investment theory) is worse than no proxy at all because it provides false confidence in a wrong answer.conf: highextends ←TBM 437: Tokens, Hours, Points, and Other Curious Proxies · p6 — Generalizes Cutler's own earlier 'a bad proxy is worse than no proxy' point from story points specifically to the full lineage of proxies (hours, flow metrics, tokens) sharing the same failure mode.
- explicitdevelopedThe appropriate epistemic framework for engineering ROI is Bayesian updating -- starting with priors, collecting signals, narrowing uncertainty quarterly -- not demanding definitive proof.conf: highextends ←TBM 437: Tokens, Hours, Points, and Other Curious Proxies · p13 — Extends Cutler's Bayesian-updating framing for engineering ROI into a concrete template of falsifiable 'theory of the system' hypothesis statements specifically for AI spend.
- explicitSuccessful engineering teams operate more on conviction backed by signals than on precise ROI calculations.conf: high
- explicitHiring discipline (being 'incredibly stingy') is a forcing function that compels teams to explore all paths before adding headcount.conf: high
- explicitRevenue per engineer is a statement about the overall business model, not about engineering efficiency.conf: high
- implicitdevelopedOrganizations that invest in codebase maintenance ('gardening') compound their advantage and avoid the entropy that cripples peers.conf: mediumextends ←TBM 430: Incubate, Compound, Refinance, Liquidate · p10 — TBM 416 claims organizations that invest in codebase 'gardening' compound their advantage over peers; TBM 430's implicit claim about intentional-movement organizations compounding faster under AI generalizes this from codebase maintenance specifically to portfolio-movement discipline broadly.
- implicitThe fact that PE acquisitions produce clarity about investment models implies that most companies' existing models are dishonest or absent.conf: medium
Arguments
Engineering ROI cannot be cold-started; it must be built as an ongoing habit. strength: high
Explicit premises:
- Most companies only ask the ROI question after years of dysfunction
- The data and relationships accumulated under the old regime don't carry over when you change regimes (Lucas Critique)
- By the time finger-pointing starts, there's no admission of responsibility and no foundation for measurement
Implicit premises:
- Organizational measurement requires trust, shared language, and accumulated context that can only be built incrementally
- The act of measuring changes what is measured (reflexivity), so measurement must co-evolve with the organization
Inferences:
- Therefore, attempting retroactive ROI measurement after years of neglect will produce misleading results
- Therefore, the only reliable path is to build stewardship habits from the start
The appropriate epistemic approach to engineering ROI is Bayesian updating, not precise calculation. strength: medium
Explicit premises:
- Product work is several causal hops from revenue
- You can never fully eliminate uncertainty in the causal chain
- A good-enough directionally correct proxy is genuinely valuable
- Successful teams operated on conviction and signals, not precise math
Implicit premises:
- Bayesian reasoning is the correct framework for decisions under irreducible uncertainty
- Organizational decision-making can meaningfully approximate Bayesian updating even without formal statistical methods
Inferences:
- Therefore, demanding precise ROI is a category error (frequentist fallacy)
- Therefore, the goal is narrowing uncertainty over time through quarterly evidence collection
Hiring discipline is a core component of investment stewardship. strength: medium
Explicit premises:
- Money is the easiest part; adding people rarely fixes a team and can obscure the actual issue
- Asking for more people implies everything else is more important and that no efficiency gains exist
- Increasing team size without exploring all paths is the real problem
Implicit premises:
- Organizational complexity grows super-linearly with headcount
- The cost of a bad hire or unnecessary hire compounds over time through coordination costs
Inferences:
- Therefore, being stingy about hiring while supporting existing teams is a superior default
Narratives
- A company with 2,000 engineers wakes up one morning and tries to cold-start the question of what the engineers are worth, having never tracked outcomes or leading indicators.
→ Illustrates the impossibility of retroactive ROI measurement; makes the abstract cold-start problem vivid and recognizable - An executive asks 'Why can't we be like Amazon?' and the author responds that you'd need to rewind 20+ years and make operational efficiency a core trait from the beginning.
→ Supports the thesis that stewardship is a long-term habit, not something you can adopt overnight; illustrates the regime-change problem - Private equity acquisition as a moment of clarity: PE firms arrive with explicit models, EBITDA targets, and clear expectations for every dollar of investment.
→ Concession narrative -- acknowledges that PE's reductive approach at least provides the honesty that was previously missing, which implicitly indicts the status quo at most companies - An engineering leader's three observations about money, stewardship, and headcount requests.
→ Provides authority and concreteness to the argument that hiring is the wrong default response; introduces the stewardship-of-capital framing - Teams that 'garden their codebase' still shipping effectively two years later while peers struggle.
→ Illustrates the compounding nature of stewardship habits and provides a concrete example of the flywheel effect
Questions addressed
- ✅ What are we getting from our engineering investment?
This question is reframed: it cannot be answered as a one-time calculation but only through continuously practiced stewardship behaviors that build the foundation for answering it over time. - ✅ Why do attempts to measure engineering ROI typically fail?
They fail because companies ask too late (after years of dysfunction), use bad proxies, face a regime-change problem where old data is invalid, and get trapped in a false binary of perfect precision vs. nothing. - ✅ What should organizations do instead of trying to calculate engineering ROI?
Build ongoing habits: identify leading indicators, build a business model, do quarterly Bayesian updates, apply hiring discipline, and have honest conversations with finance. - ⭕ How do you know if a team is delivering good ROI without precise measurement?
Partially addressed through the concept of conviction plus signals and good-enough proxies, but the specific criteria for distinguishing genuine conviction from self-deception are not explored.
Gaps
- incompleteThe article doesn't address the organizational power dynamics that make stewardship habits hard to sustain. Who has the authority to insist on these practices? What happens when a new VP arrives and disrupts the habit? The argument assumes a continuity of leadership and culture that many organizations lack.
Read next: Literature on organizational routines and institutional memory (e.g., Nelson & Winter's evolutionary theory of the firm) would strengthen this argument. - incompleteThe Bayesian updating framework is invoked but never operationalized. What are the priors? What counts as evidence? How do you avoid anchoring bias in your priors? The metaphor does real work in the argument but remains at the level of analogy.
Read next: Douglas Hubbard's 'How to Measure Anything' provides practical methods for Bayesian estimation in business contexts that could concretize this. - incompleteThe article conflates two distinct audiences: engineering leaders who need to practice stewardship, and executives/finance who need to ask better questions. The five action steps mix both perspectives without differentiating who owns each behavior.
Read next: Exploring the principal-agent problem between engineering and finance would sharpen the argument about whose habit this needs to be. - incompleteThe PE aside is presented as a concession but actually undermines part of the thesis. If PE firms can arrive and quickly create clarity through explicit models, that suggests some form of cold-start IS possible -- just a reductive one. The author waves this away without fully engaging the tension.
Read next: A more careful treatment would distinguish between types of clarity: PE provides cost-side clarity but not value-side clarity, which is the harder problem the article is actually about.