Odd Lots2026.06.191 hr 6 min

Inside Anthropic's economic research on recursive AI scaling and market disruption

Original title · Anthropic's Co-Founder and Top Economist on Doing Research at the AI Frontier

用阿法問答深入這集 ✨

或自己問：

US retail investors eye AI stocks for raw power and computing scale, but the real structural shift is occurring inside the development labs themselves. At Anthropic, the creators of the Claude LLM, the speed of internal software engineering has spiked by approximately eight times over a multi-year run rate because developers are letting AI agents write and integrate their own code. This compounding cycle of "recursive self-improvement" hints at a fast-approaching productivity shock that has yet to register in backward-looking macro data, presenting both a massive margin tailwind for tech adopters and a sudden bottleneck for entry-level white-collar employment.

Internal compounding outpaces public economic metrics

The disconnect between skyrocketing AI capabilities and sluggish national productivity data is a classic lagging indicator. While macro economists struggle to find the AI signal in noisy post-pandemic GDP revisions, internal metrics at frontier labs show an exponential curve. Anthropic's developers are deploying multiple coordinate coding agents to automate raw engineering tasks, resulting in an 8x surge in code output. For investors, this suggests that productivity gains will initially consolidate inside highly integrated, tech-native firms before diffusing into the broader S&P 500.

The white-collar labor market faces a barbell risk

A structural shift in hiring patterns is emerging as AI tools automate execution-heavy tasks. Companies are shifting toward a "barbell" hiring model, heavily favoring highly experienced specialists with deep domain expertise who can direct and audit AI agents, while cutting back on entry-level roles. Early career workers in highly exposed cognitive roles are already showing lower job-finding rates. This is easy to overlook in a strong overall labor market, but it signals that the traditional corporate pipeline of training junior staff to do rote analytical work is permanently breaking down.

Trust is the ultimate enterprise moat

While the market frequently rewards raw computing speed and parameter size, enterprise buyers prioritize safety, reliability, and strict risk guardrails. This dynamic mirrors the automotive industry, where premium safety engineering coexists with high performance to drive consumer adoption. For long-term investors, the real commercial winners won't just be the ones running the fastest models, but the platforms that build robust auditing systems, prevent data leakage, and establish institutional trust.

The critical signal to watch next is how successfully large, legacy enterprise customers restructure their internal firewalls to feed proprietary context into these frontier models.

Assets mentioned in this episode

NVDANvidia▲ 看多
The host points to Nvidia as the fundamental hardware layer of the current AI boom, reflecting on his failure to publish a timely story on its ubiquitous research presence prior to leaving journalism in 2016. The underlying investment logic rests on the compounding compute demands of continuous model scaling, where every increase in neural network resources yields superior emergent properties. This hardware-driven cycle serves as the prior analog to the software-driven recursive loop currently being tested.

Key questions

Why is the traditional corporate apprenticeship model breaking down?→

Routine tasks like data cleaning and basic coding, once used to train junior staff, are now automated. Consequently, entry-level workers face weaker job-finding rates as firms shift toward lean, highly leveraged teams of elite experts.

💡 詢問阿法，看詳細逐字稿分析→

What is the biggest barrier to capturing the potential AI productivity dividend?→

The primary bottleneck is not model intelligence, but disorganized, siloed corporate data. Only firms with modernized, clean, and highly structured data pipelines can effectively deploy agentic workflows to unlock significant cost savings.

💡 詢問阿法，看詳細逐字稿分析→

How is the developer economy shifting under the weight of AI-generated code?→

Developers are now writing significantly more code by directing autonomous agents, shifting the bottleneck to validation. Competitive advantage now belongs to firms that build robust, automated systems to stress-test synthetic output rather than reviewing it manually.

💡 詢問阿法，看詳細逐字稿分析→

Further research

Tickers and signals often linked to this episode's themes in public sources · AI-compiled, not investment advice

Data Architecture as the AI Bottleneck

Enterprise AI ROI is currently gated by clean data pipelines, making companies with modernized, centralized data architecture the primary beneficiaries before AI adoption can scale.

US stocks

SNOW
SnowflakeBenefitsProvides a unified cloud data platform that centralizes enterprise data silos, making it the bedrock for clean, structured data pipelines required for AI model scaling.
PLTR
Palantir TechnologiesBenefitsDelivers the foundational semantic layer and AIP system that integrates fragmented enterprise data structures into a cohesive model for contextual AI orchestration.
MDB
MongoDBBenefitsSupplies flexible document-model database structures that are crucial for managing the unstructured and semi-structured data feeding AI search and context engines.

Risks

A prolonged stagnation in enterprise IT budgets or the rapid commoditization of cloud-native data streaming tools could delay the execution of centralized data modernization pipelines.

Watch list

Quarterly product revenue growth at Snowflake
Enterprise customer adoption and implementation metrics for Palantir AIP
Database migration trends from legacy SQL databases to flexible document-model structures

AI Audit and Safety Compliance Industry

As models gain autonomous capabilities, the shift toward mandatory third-party safety testing and KYC for compute clusters creates a predictable, lucrative demand for an emerging AI compliance and certification sector.

US stocks

VRNS
Varonis SystemsBenefitsOffers an AI-native Data Security Platform and the Atlas framework specifically designed for continuous AI compliance, data security posture management, and least-privilege automation.
ZS
ZscalerBenefitsIntegrates automated AI red-teaming, runtime compliance, and safety threat analysis to secure generative AI models throughout their entire enterprise deployment lifecycle.
S
SentinelOneBenefitsFeatures specialized Prompt AI Red Teaming software designed to let enterprise security teams aggressively test and fortify their custom-built AI models against adversarial prompt injections.

Risks

A regulatory rollback or dilution of mandatory AI safety laws could reduce the legal urgency for enterprises to purchase commercial AI red-teaming and compliance software.

Watch list

Enforcement timelines of the EU AI Act and regional US AI safety legislation
Enterprise adoption rates of security suites containing AI runtime guardrails
Quarterly growth in security SaaS bookings for automated AI compliance products

Corporate Capex Shift: Hardware vs. Process Engineering

The sustainability of the current AI cycle depends on enterprise capital expenditure moving beyond simple GPU procurement into costly internal process reengineering and data readiness, which serves as a leading indicator of actual economic diffusion.

US stocks

ACN
AccentureBenefitsBenefits from massive consulting and deployment bookings as enterprises shift their focus from raw hardware procurement to costly integration and operational process redesign.
PATH
UiPathBenefitsOrchestrates autonomous agents and redesigns legacy enterprise workflows with its process intelligence and human-in-the-loop agent testing platforms.
NVDA
NvidiaPressuredFaces potential capex digestion and valuation pressure if enterprise spend rotates away from aggressive GPU chip procurement toward software readiness and consulting services.

Risks

If the economic ROI of initial AI process re-engineering projects disappoints, enterprises may freeze overall AI initiatives entirely, hurting both hardware and service providers.

Watch list

Accenture's advanced AI bookings and general consulting revenue growth
Enterprise IT budget allocation splits between hardware/GPUs and IT services/consulting
Adoption and integration metrics of agentic process automation platforms like UiPath

This section is AI-compiled from public sources, may be inaccurate or outdated, is for research reference only, and is not investment advice.

Want the full episode?