The Unjournal · Research prioritization
Paper-specific consideration page · Top AI-related papers, rank 2

Canaries in the Gold Mine: Early Productivity Gains from Artificial Intelligence Creating Organization Capital

Think-tank use of the headline result; no extended scrutiny found
Why this page exists. This paper is on a shortlist of papers we are considering for independent evaluation. No human ratings have been submitted yet, so the score below comes from the AI scoring pass alone. Treat it as one provisional input; the team has made no prioritization decision.
65AI evaluation priority · later Opus assessment
70
Earlier AI score (comparison only)

The combined score uses the later assessment shown alongside. This earlier score is retained for comparison.

None yetHuman aggregate · no ratings submitted
Not availableHuman–AI synthesis needs at least one human rating

Human input

No human ratings have been submitted for this paper yet, so there is no human aggregate or synthesis score. If you know the work, rate or discuss it on the dashboard card; the team reviews feedback before it affects prioritization.

Loading public input and its source labels…

AI assessment: why consider this paper?

The paper asks whether recent AI investment is already raising firm productivity, and through what channel. It reports that firms with more AI-skilled employment saw faster productivity growth in recent years but not in the prior decade, and attributes this to organization capital measured from job descriptions. An evaluation would focus on identification (AI-skilled hiring is endogenous), the novel organization-capital measure, and whether the result generalises beyond listed or well-measured firms.

AI-generated criterion ratings and reasoning

Scores come from AI prioritization; the accompanying explanations include AI-assisted source checks. These are provisional judgments, separate from human ratings and commissioned evaluations.

Decision relevance

6.0/10

The research could improve choices about AI investment and organizational complements, and help distinguish persistent productivity gains from changes that should not alter macroeconomic expectations. Those choices affect workers, consumers and owners, potentially across countries, but the paper's incremental welfare contribution is much smaller and less certain than the total stakes of AI adoption. Its useful addition is firm-level evidence and a proposed mechanism alongside existing experiments, surveys and organizational research. Distribution, implementation costs, displacement and actual changes in decisions remain unresolved. LMIC benefits could be especially valuable through income, service provision and new local evidence, but transfer is unproven; animal-welfare and catastrophic-risk pathways are speculative. The impact score therefore reflects a moderate positive research-VoI case, independently of whether Unjournal should evaluate it.

  • Paper claim to check The authors report a positive relationship between internal AI investment and firm productivity growth during 2018–2024, unlike the earlier period. Their conference abstract also reports instrumental-variable estimates using historical hiring ties to AI-intensive universities as support for a causal interpretation. [NBER conference abstract](https://www.nber.org/conferences/firms-technology-and-economic-implications-fall-2026) Source abstract

Value of added scrutiny

7.0/10

The NBER paper (August 2026) has been picked up by the Information Technology and Innovation Foundation, a think tank, which used its headline finding. We found no replication or extended critique. Attention looks real but early.

Timing

7.5/10

The paper was released in August 2026 and remains listed as a working paper; the current author page reports a Journal of Financial Economics revise-and-resubmit. Feedback can still inform revision, but this is not the maximum-value case of a wholly unreviewed manuscript. [CEPR record](https://cepr.org/publications/dp21894) [Author's research page](https://www.alexxihe.com/) Additional checking found Jose Maria Barrero's publicly listed September 2026 NBER conference discussion of this exact paper. Its linked slides could not be retrieved, so their claim coverage, substantive findings and correspondence to the current revision are unknown. [Discussion listing](https://www.jmbarrero.com/discussions) This establishes a relevant scrutiny lead, not comprehensive validation. Recent Federal Reserve citation makes a timely public clarification useful; discussion and attention should not mechanically reduce urgency.

  • Source record Publication-stage and date evidence should be checked in the linked paper record. NBER

Methodological potential

7.0/10

The public repository provides construction code, AI keywords and synthetic inputs, but excludes the actual firm panel and proprietary underlying data; it is not a complete publicly executable replication of the headline results. [Author repository](https://github.com/jrhao7/AI_productivity_replication) An evaluator would need access sufficient to examine firm matching, sample selection, worker coverage and classification stability. Identification requires assessing whether historical university hiring links affect productivity through other skills, innovation or networks, alongside instrument strength and appropriate inference. Measurement review should test whether the AI and organization-capital classifications capture distinct constructs and whether changes in descriptions create artificial trends. Mechanism assessment must distinguish evidence consistent with mediation from identified mediation; different significance levels across job categories alone do not establish different effects. Productivity interpretation also requires checking prices, markups, intangible inputs and aggregation. A narrow evaluation should prioritize these consequential claims, reconcile accessible prior criticism and state which conclusions remain untestable without data. This prioritization pass did not inspect the full manuscript or reproduce its estimates.

  • Paper claim to check The authors report a positive relationship between internal AI investment and firm productivity growth during 2018–2024, unlike the earlier period. Their conference abstract also reports instrumental-variable estimates using historical hiring ties to AI-intensive universities as support for a causal interpretation. [NBER conference abstract](https://www.nber.org/conferences/firms-technology-and-economic-implications-fall-2026) Source abstract

Prominence

9.0/10

The scoring model estimated prominence from the paper's venue, authors, institutional setting, and visibility. The model did not supply a criterion-specific explanation. Current public-attention status: Think-tank use of the headline result; no extended scrutiny found.

  • Think-tank use ITIF highlighted the paper's main estimate in a short public note. This shows the result is circulating in policy-facing commentary, though the note is brief and does not appraise the identification. ITIF

Likely influence

9.0/10

The scoring model estimated how far the findings could shape later research or decisions, without supplying a criterion-specific explanation. Current public-attention status: Think-tank use of the headline result; no extended scrutiny found. See public-attention evidence below.

  • Think-tank use ITIF highlighted the paper's main estimate in a short public note. This shows the result is circulating in policy-facing commentary, though the note is brief and does not appraise the identification. ITIF

What the paper says

Source abstract · Source-supplied abstract wording; HTML entities and whitespace normalized. Not independently compared with the paper PDF.

Using a new firm-level measure of AI investment based on AI-skilled employment—spanning machine learning through generative and agentic AI—we show that AI investments are associated with productivity growth in recent years, but not over the previous decade. We trace the productivity gains to the accumulation of organization capital that AI helps create: durable firm-specific knowledge acquired through learning-by-doing that enables more efficient production. We build a novel measure of organization capital based on workers’ job descriptions and document that productivity gains are driven by AI-skilled jobs that build organization capital. Overall, our findings suggest that AI investment generates productivity growth by creating organization capital.

Claims to check

  • The authors report a positive relationship between internal AI investment and firm productivity growth during 2018–2024, unlike the earlier period. Their conference abstract also reports instrumental-variable estimates using historical hiring ties to AI-intensive universities as support for a causal interpretation. [NBER conference abstract](https://www.nber.org/conferences/firms-technology-and-economic-implications-fall-2026)
  • A policy-facing summary attributes to the paper 7.5 percentage points more total-factor-productivity growth and 6.2 percentage points more sales-per-worker growth over six years for a one-standard-deviation greater increase in AI-skilled worker share. These circulating magnitudes are evaluation targets, not independently verified table estimates; the exact sample, scaling and uncertainty require checking. [Technology Policy Institute summary](https://techpolicyinstitute.org/publications/economics-and-methods/research-roundup/research-roundup-october-2026/)
  • The paper attributes productivity gains to AI-skilled jobs classified as building organization capital, using workers' job descriptions to measure that activity. The consequential claim is that this pattern identifies a mechanism through durable firm-specific knowledge, rather than merely correlated hiring or organizational change. [CEPR abstract](https://cepr.org/publications/dp21894)
Methodological or theoretical issues flagged for evaluation

The public repository provides construction code, AI keywords and synthetic inputs, but excludes the actual firm panel and proprietary underlying data; it is not a complete publicly executable replication of the headline results. [Author repository](https://github.com/jrhao7/AI_productivity_replication) An evaluator would need access sufficient to examine firm matching, sample selection, worker coverage and classification stability. Identification requires assessing whether historical university hiring links affect productivity through other skills, innovation or networks, alongside instrument strength and appropriate inference. Measurement review should test whether the AI and organization-capital classifications capture distinct constructs and whether changes in descriptions create artificial trends. Mechanism assessment must distinguish evidence consistent with mediation from identified mediation; different significance levels across job categories alone do not establish different effects. Productivity interpretation also requires checking prices, markups, intangible inputs and aggregation. A narrow evaluation should prioritize these consequential claims, reconcile accessible prior criticism and state which conclusions remain untestable without data. This prioritization pass did not inspect the full manuscript or reproduce its estimates.

Opus 5.5 re-evaluation (experimental)

Opus raw score: 60/100
Adjusted (+15): 75/100
Opus own action: watchlist
Label from adjusted score: shortlist
Earlier score (gpt-5.5): 80/100
Shift applied to the AI score: -5

Read with care. This is an experimental re-evaluation by a different model (Claude Opus 5.5, 2026-10-01). Across a calibration sample Opus scored about 15 points lower than GPT-5.5, so 15 points are added to compare it with GPT-5.5 scores. The calibration is a small sample (n=44) and is not human-validated, and the two models disagree about the order of papers within the top group. The AI score at the top of this page is the usual evaluation-priority score shifted by the difference between the adjusted Opus score and the earlier holistic score; the synthesis uses that shifted value, and the unshifted score is shown beside it. On the dashboard you can switch the Opus scores off. Read the methods note.

Opus rationale (AI-generated)

This NBER working paper by Babina, He and Jiang extends the authors' well-known line of work, using AI-skilled employment as a firm-level measure of AI investment, to argue that AI investment now goes with productivity growth (unlike the previous decade), and that it does so by building organisation capital. That speaks directly to one of the most active economic-policy debates: whether and when AI raises productivity, and what complementary investments are needed. Fed, OECD, IMF and CBO productivity forecasters, AI-policy think tanks (Brookings, GovAI, Epoch AI) and firms deciding how to deploy AI all care about it. It is squarely in scope for Unjournal and still an unrefereed working paper, so evaluation would be timely. An evaluation could add value by scrutinising identification (selection of AI-investing firms, endogenous timing), whether the new text-based organisation capital measure is valid, and how productivity is measured, since these determine whether the 'canaries' framing is warranted. Concerns: the welfare pathway runs through rich-country firm productivity, with little direct LMIC, animal or catastrophic-risk relevance. The evidence is probably correlational, and many groups are already working on the AI-productivity question. I'd put it in the monitor zone, near the upper end, and suitable if Unjournal wants AI-economics coverage.

Dashboard details and provenance

Discovery source: NBER
Scoring model: gpt-6-astra (codex headless, xhigh)
Model holistic score: 72.0

Full AI dashboard scoring rationale

This belongs near the top of the monitoring list: it could inform whether businesses fund organizational changes alongside AI adoption and how policymakers interpret emerging productivity evidence. The Federal Reserve has already cited the paper in a policy speech, while the OECD and World Bank are plausible users of its implications for diffusion and complementary investment; no changed policy or funding allocation was verified. [Federal Reserve speech](https://www.federalreserve.gov/newsevents/speech/jefferson20261113a.htm) It remains a working paper, but an author's current page lists a Journal of Financial Economics revise-and-resubmit, so describing it as having received no independent review would be misleading. [Author's research page](https://www.alexxihe.com/) A focused public evaluation could assess the causal interpretation, validity of the organization-capital measure and limits of aggregate extrapolation, making explicit which conclusions policy readers can use. I would favor that bounded commission over an unrestricted replication, after mapping the existing conference discussion and confirming access: the welfare contribution is moderately promising, existing scrutiny is incompletely observed, and unavailable underlying data could make a full evaluation costly.

AI decision-relevance rationale

The research could improve choices about AI investment and organizational complements, and help distinguish persistent productivity gains from changes that should not alter macroeconomic expectations. Those choices affect workers, consumers and owners, potentially across countries, but the paper's incremental welfare contribution is much smaller and less certain than the total stakes of AI adoption. Its useful addition is firm-level evidence and a proposed mechanism alongside existing experiments, surveys and organizational research. Distribution, implementation costs, displacement and actual changes in decisions remain unresolved. LMIC benefits could be especially valuable through income, service provision and new local evidence, but transfer is unproven; animal-welfare and catastrophic-risk pathways are speculative. The impact score therefore reflects a moderate positive research-VoI case, independently of whether Unjournal should evaluate it.

AI timing assessment

The paper was released in August 2026 and remains listed as a working paper; the current author page reports a Journal of Financial Economics revise-and-resubmit. Feedback can still inform revision, but this is not the maximum-value case of a wholly unreviewed manuscript. [CEPR record](https://cepr.org/publications/dp21894) [Author's research page](https://www.alexxihe.com/) Additional checking found Jose Maria Barrero's publicly listed September 2026 NBER conference discussion of this exact paper. Its linked slides could not be retrieved, so their claim coverage, substantive findings and correspondence to the current revision are unknown. [Discussion listing](https://www.jmbarrero.com/discussions) This establishes a relevant scrutiny lead, not comprehensive validation. Recent Federal Reserve citation makes a timely public clarification useful; discussion and attention should not mechanically reduce urgency.

Earlier public-scrutiny search

The supplied search does not establish substantial expert scrutiny: most hits are unrelated, and the Cameroon article is a downstream citation lead describing a different empirical study. Its full text could not be checked, so it cannot be credited as a replication or methodological assessment. Empty annotations cover only the queried URL variants. Additional primary-source checks, documented in timing_rationale, establish a public conference-discussion listing and an author-reported journal revise-and-resubmit; the discussion content and referee reports were unavailable. Thus the paper is neither demonstrably untouched nor demonstrably thoroughly vetted. The evaluation-value score reflects documented policy-facing use and potentially consequential unresolved interpretation, tempered by uncertain duplication and data access. A feasible commission should map existing discussion against the current manuscript, then assess identification, measurement and the organizational mechanism, with targeted reproduction if access permits. A public claim assessment could explain which findings support business or forecasting decisions and which extrapolations remain unsupported. Merely running the released construction code would not validate those claims.

  • https://doi.org/10.1177/2319510x261494732

Intake, review, and crux connections

Community crux

Do general-purpose technological revolutions reliably precede (and sometimes temporarily depress) the measured productivity growth they eventually cause? · 70% match

Firm evidence on AI-driven organization capital directly informs whether GPTs produce delayed productivity gains through complementary investment.

Community crux

Will AI's transformative narrative impacts actually show up in 2030 macro variables like labor share, or will diffusion/infrastructure barriers keep them modest? · 52% match

Early firm productivity effects help assess whether AI impacts will appear in near-term macro variables or remain diffusion-limited.

Public attention and use

AI-assisted review of the linked public sources.

Research dissemination listing / discoverability

NBER working paper 35684 and SSRN listing

The paper is available through NBER and SSRN. These improve discoverability but are not independent scrutiny.

Source: SSRN · Relationship: bibliographic index

What the search did not establish

  • No exact-title EA Forum or LessWrong discussion surfaced in the targeted search.
  • No replication, comment paper or detailed methodological critique was found.
  • The most decision-relevant next signal is independent re-estimation using a different measure of AI investment.

Targeted search checked 2026-10-01. Search scope: Targeted exact-title searches across the open web, institutional and author pages, news and public social-media results, EA Forum/LessWrong, and policy/white-paper contexts. Evidence records distinguish commissioning or report use from independent discussion, media attention, indexing, and post-publication policy use. A search miss is reported as uncertainty, not proof of absence. Entries with check_depth "quick" rest on roughly one to two searches and should be read as especially uncertain; entries checked on 2026-10-01 were added for the shortlist of papers without human ratings.