● Google Soars as AI King- Antitrust Foiled
Google Crowned the Strongest Player in the AI Revolution.. This is What Happens Next
Key contents covered in this article — essential points you must read:
- Today’s Verdict’s Core: Ruling against the forced sale of Chrome and Android, and alleviation of antitrust risk
- Immediate Market Reaction: Google +9%, Apple +4%, Nasdaq +1% – a ‘top 2 hard carry’ mechanism
- True Meaning of the Ruling: The decisive impact of generative AI (ChatGPT, Gemini, etc.) on the monopoly power judgment
- Practical Issues of Data Sharing: What data, to whom, and how it is likely to be transferred
- Investment Perspective Summary: Valuation, analyst reactions, portfolio checklist
- Future Scenarios and Risk Management (Appeal possibility · Data restrictions · Intensified competition)
- Practical Tool Recommendation: How to use InvestingPro and my discount code
1) Event Timeline — Key News (Reflected Immediately) Today
In the antitrust lawsuit between the Department of Justice and Google, the court ruled that there was no need to force Google to sell its Chrome web browser or Android.
Immediately after this ruling, Google’s stock price rose by approximately 9%, and Apple’s surged by about 4%, with the sharp increase in these two stocks driving the overall Nasdaq (+1%) rise.
The judge explicitly stated that the emergence of generative AI has changed the market landscape, and this point served as a key variable in the ruling.
2) Key Points of the Ruling — Antitrust, Data Sharing, and the Role of AI
Key 1: The court believed that competition could be improved without the forced sale of assets such as Chrome and Android.
Key 2: The payments Google makes to smartphone manufacturers to provide search as a default on iOS and Android devices (typically large ‘default deals’) were not entirely prohibited.
Key 3: The court requested search data sharing and enhanced competitor access to promote competition.
Key 4: The judge determined that Google’s monopolistic power has weakened as generative AI (e.g., ChatGPT, Gemini, etc.) emerged as an alternative channel to traditional search.
3) Data Sharing: Who Will Receive What — Practical Issues
The court’s demand for ‘data sharing’ appears simple in a short sentence but is actually very complex.
What will be shared: It is highly likely that aggregated, anonymized, or sampled search trends and usage patterns will be central, rather than full search logs.
Who will receive it: US-based AI companies (e.g., OpenAI, xAI, Anthropic, etc.) are primary candidates.
Legal and Technical Limitations: Due to personal identifiable information (PII) protection, GDPR-like regulations, corporate security, and trade secret issues, a ‘complete’ data transfer is highly unlikely to be feasible.
Therefore, the realistic scenario involves ‘partial, restricted, and anonymized’ data sharing, which would not immediately dismantle Google’s moat.
4) AI Competitive Landscape — Gemini, GPT, Shift in Market Share
As the judge explicitly stated, generative AI is transforming search behavior itself.
According to reported user numbers (industry estimates), there are talks of ChatGPT having around 800 million users and the Gemini family around 500 million, which are estimates considering market impact.
With Gemini 2.5 significantly boosting its performance level, Google’s AI presence has been reconfirmed, and anticipation for Gemini 3 is growing.
According to research rumors from sources like SemiAnalysis, Gemini 3 is expected to show significant improvements in multimodal and coding capabilities.
However, the ultimate winner is not determined solely by model performance.
Data accessibility, infrastructure (cloud), ecosystem (apps·plugins), and regulatory risks must all be considered.
5) Market Reaction and Valuation (Investment Perspective)
The market reacted immediately as Google’s benefit from the ruling appeared clear in the short term.
Analyst coverage is also being rapidly updated, and the currently announced target prices (e.g., $217) are mixed figures, reflecting both pre- and post-ruling sentiments.
Google’s forward P/E ratio is relatively low among big tech companies (19x by the standard you used), highlighting its value appeal.
Comparison: It’s worth noting that Meta is at 23x, while MS/Amazon/Apple are in the mid-30s.
Investment Check: Considering the immediate alleviation of regulatory risk and AI growth momentum, an overweight position may be attractive, but appeal and detailed conditions for data sharing remain variables for a rerating.
6) Practical Investment Strategy (Recommended Actions and Monitoring Items)
1) Positioning: Maintain or slightly increase long-term positions (assuming there is valuation room).
2) Staggered Buying: Divide additional purchases based on the launch of Gemini 3, the outcome of legal appeals, and detailed agreements on data sharing.
3) Hedge: Hedge some risk with options or related infrastructure (cloud) stocks in preparation for regulatory worsening.
Monitoring Points: (A) Court appeal status and timeline, (B) Scope of data sharing (sampling, aggregation, anonymization status), (C) Gemini 3 announcement and performance metrics, (D) Potential disclosure of default deal sizes by manufacturers like Apple and Samsung.
7) Risk Scenarios Summary
Best Scenario: No appeal filed · Data sharing at a limited level, Google strengthens market dominance with AI competitiveness.
Base Scenario: Appeal filed but prolonged, partial data sharing, Google stabilizes after short-term volatility.
Worst Scenario: Appeal dismissed, followed by detailed mandatory measures (widespread data sharing · platform separation), weakening long-term growth momentum.
8) Wall Street and Analyst Trends
Coverage: A majority of analysts are updating their reports immediately after the ruling.
Initial target prices (e.g., $217) are merely temporary benchmarks, and upward revisions through rapid reratings are highly likely to follow.
Sell opinions are few, with most maintaining a ‘Buy’ stance.
9) Data Beneficiaries and the AI Startup Ecosystem
The companies set to receive data are major AI firms both domestically and internationally, and their improved model performance will soon lead to a restructuring of the search and conversational AI markets.
However, the actual quantity and quality of data will be limited, and it will be difficult for new startups to replace Google overnight.
10) Tool Recommendation — InvestingPro (Including Promotion)
For stock research, financial indicators, and tracking expert portfolios, InvestingPro is useful for practical investors.
Disadvantages include that it is not an all-weather coverage like Bloomberg, and backtest results may have survivorship bias.
If you need it, an additional 15% discount will be applied when you register for Pro using my discount link.
InvestingPro Additional 15% Discount
Inquiries: waltechguy@gmail.com
11) Practical Conclusion (Summary Action Guide)
Google’s short-term stock price surge reflects expectations of eased antitrust risk.
The rise of generative AI signifies a structural change in the search market, which will act as both an ‘opportunity’ and an ‘increase in competitors’ for Google in the long term.
The effectiveness of data sharing is likely to be limited, so Google’s data moat and technological/infrastructure advantages will not disappear immediately.
Investors should use valuation room for staggered buying, monitor legal and technical events (appeals, Gemini 3), and employ hedging strategies against risks.
This ruling did not force Google to sell Chrome or Android, and the rise of generative AI played a key variable in the court’s decision.
In the short term, the market reaction led by Google and Apple was significant, while in the mid-to-long term, the scope of data sharing and the AI competitive landscape are key.
Investors should reasonably combine staggered buying, monitoring (appeals/data specifics/Gemini 3), and hedging strategies.
[Related Articles…]
Google: Will it be a Winner in the Generative AI Competition? Key Strategy Analysis
Impact of AI Regulations and Antitrust Rulings on the Market
*Source: [ 월텍남 – 월스트리트 테크남 ]
– AI혁명의 최강자로 등극하는 구글.. 앞으로 “이렇게” 됩니다
● Model T vs. BlackBerry Moment – Ford’s EV Gamble, Apple’s AI Crisis
What do Apple and Ford have in common now? Model T Moment vs. BlackBerry Moment — Including 7 Key Insights
This provides an at-a-glance summary of the commonalities and differences between Ford and Apple, and what turning points lie ahead.The three most important points, not well covered in other news, are as follows:1) Ford’s ‘factory structure redesign’ is not just about cost reduction but a strategic gamble to change its dependence on supply chains, batteries, and software.2) Apple’s real risk is not ‘lack of AI technology’ but ‘slowdown resulting from a combination of insufficient ecosystem openness and talent outflow’.3) For both companies, a single innovation (hardware, algorithm, or production method) alone will not secure victory; simultaneous improvements in ‘ecosystem, factory, and talent’ are necessary.
1) Historical Timeline — From Model T to BlackBerry (1908 → 2000s)
The Model T’s popularization of automobiles through mass production and standardization in 1908 marked a paradigm shift in the industry.The conveyor belt, standardized parts, and single-color strategy drastically lowered prices, exploding demand.In the 2000s, BlackBerry led the smartphone market with software and services like corporate email.However, during the transition to touchscreens and app ecosystems, its market share plummeted due to hardware obsession and failure to adapt to the ecosystem.These two cases offer a common lesson: ‘choices at inflection points’ determine the survival or demise of a company.
2) Ford’s Present (Timeline, Strategy, Numbers)
In 2021, Ford announced a major transition to electric vehicles (EVs) and promised large-scale investments.However, in 2024, Ford admitted that its plan for a full EV transition in Europe was overly ambitious.Ford’s EV business has recorded continuous losses since 2022, with estimated losses per vehicle being very significant.The price gap between Chinese EVs (average approximately $24,000) and the average US EV price (approximately $55,500) increases competitive pressure.Ford’s proposed solution is a ‘universal EV platform’ and a ‘universal EV production system’.This system aims to reduce the number of processes and parts and increase assembly speed by transforming the assembly line into a tree (module) structure.Expected effects include quantitative targets such as a 40% reduction in processes, a 20% reduction in parts, and up to a 15% improvement in assembly speed.This strategy is a modern reinterpretation of the Model T’s mass production philosophy, with securing price competitiveness as its core goal.However, key risks include battery costs, software platform competitiveness, and price/supply chain competition with Chinese OEMs.For actual success, a decrease in battery cost per kWh, stabilization of production module quality, and securing software (OTA, EV UX) must be pursued in parallel.
3) Apple’s Present (Timeline, Strategy, Numbers)
Apple is a company that created the ‘iPhone moment’ with the iPhone, but currently, it is facing ‘BlackBerry moment’ concerns in the AI competition.Securities analysts like Dan Ives have pointed out a significant risk if Apple falls behind in AI development speed and talent acquisition.There have been successive instances of key personnel leaving Apple’s AI foundation model team, and schedule delays, such as the integration of Siri-based LLM being pushed back to 2026, have occurred.Market distrust grew as AI innovation expectations were not met at WWDC 2025.Apple’s strengths include a robust hardware and services ecosystem, and consumer trust built on privacy and security.However, a disadvantage is that its closed ecosystem can limit its ability to absorb rapid external innovations (models from OpenAI, Google, Meta).Apple’s possible solutions include re-securing internal talent, strategic acquisitions (e.g., external LLMs or startups), or partnerships (e.g., OpenAI, Google).Here, issues of regulation, privacy, and revenue models (App Store, services) interact complexly, making simple collaborations difficult.
4) Practical Insights Not Often Discussed by Other Media (4 Key Points)
Insight 1 — The ‘Potential Pitfalls’ of Production Tree Transformation: Modularization leads to reduced parts and processes, but if module standardization fails, production flexibility decreases, and supply chain concentration risks increase.Insight 2 — Price competition is not just about reduction but about improving ‘unit economics’. If losses per vehicle do not decrease even with increased EV sales, long-term sustainability is impossible.Insight 3 — Apple’s problem is not a technology gap but a ‘trade-off between speed and openness’. Closed ecosystems are advantageous for security and revenue but hinder progress in rapidly innovating areas like AI.Insight 4 — Both companies have a high chance of failure if they only improve a single aspect (hardware or algorithm). Simultaneous improvement of the ‘factory, software, and talent’ three pillars is necessary.
5) Key Metrics for Investors and Executives to Monitor
Ford Checklist:Quarterly trend of EV profit/loss per vehicle (loss/profit per vehicle).Battery costs (cost per kWh) and battery supply contracts (long-term, price).Number of factories converted to the universal platform and their productivity (number of processes, assembly time).Pricing and export trends of Chinese OEMs, and reactions to price competition within the US.Software revenue (subscription services, OTA update utilization rate).
Apple Checklist:AI talent recruitment and attrition status (tracking key researcher departures).Apple Intelligence (Siri LLM) feature roadmap and adherence to launch schedule.Progress of external partnerships and acquisitions.Proportion of services revenue and initial monetization potential of AI-related services.Potential product restrictions due to regulatory issues (data, privacy).
6) Outcome Predictions by Scenario (Short-term 1–2 years, Mid-term 3–5 years, Long-term 5–10 years)
Ford Conservative Scenario (Medium Probability): Continuous losses due to platform transition failure, decline in market share.Ford Neutral Scenario (High Probability): Unit cost improvement through universal platform, price gap with China remains but profitability gradually recovers.Ford Aggressive Scenario (Low Probability): Platform success and sharp drop in battery costs lead to leadership in the mass-market low-cost EV segment.
Apple Conservative Scenario (Medium Probability): AI launch delays and maintenance of closed ecosystem, stable revenue from hardware and services but loss of AI leadership.Apple Neutral Scenario (High Probability): AI capabilities enhanced through strategic acquisitions and partnerships, regaining competitiveness, albeit late.Apple Aggressive Scenario (Low Probability): Reclaiming AI ecosystem leadership through large-scale acquisitions or a major internal overhaul.
7) Conclusion — Key Message in One Sentence
Ford and Apple stand at the crossroads of manufacturing innovation and AI innovation, respectively.Neither can succeed with a single-pronged (hardware or software) approach.Only by aligning the ‘three pillars’ of production methods, software ecosystem, and key talent acquisition can they recreate a Model T Moment or an iPhone Moment.
[Related Articles…]Realistic Solutions for Reducing EV Production CostsApple’s AI Strategy Review: The Challenge Left by Talent Outflow
*Source: [ 티타임즈TV ]
– 지금 애플과 포드의 공통점은?
● Apple’s 85x Faster FastVLM Local AI Tsunami for Cloud, Chips, Software
Apple FastVLM Impact Analysis: 7 Critical Implications of 85x Faster TTFT for the Global Economy and Industry
This content provides an at-a-glance overview.What technical differentiators obliterated existing VLMs.How Apple’s local AI (MacBook engine) strategy will reshape cloud, semiconductor, and software revenue structures.10 immediate checklists for investors and businesses.Invisible ripple effects on regulations, privacy, and the labor market (key points often missed by other news).Action plans for practitioners at each stage (immediate, short-term, mid-term, long-term).(Core SEO Keywords: AI, multimodal AI, vision language model, latency, MacBook Pro)
Immediate (Post-News) — Core Technology Summary and Points Often Missed by Other Media
FastVLM’s Technical CoreApple FastVLM reduces TTFT (Time To First Token) by up to 85 times, and its vision encoder is 3.4 times smaller.It introduces Fast Vit HD, a hybrid vision encoder (convolution + transformer 5-stage structure), significantly reducing the number of visual tokens.The key point is its announced real-time operation capability on the MacBook Pro’s Neural Engine.The Most Important Point Rarely Mentioned ElsewhereApple’s very objective of designing for ‘local real-time multimodal AI’ is a fundamental strategic shift.This means performance improvement isn’t just a paper’s point but can transform ‘privacy, response speed, and cost competitiveness’ on consumer devices.This point is often overlooked by general news, which focuses solely on performance metrics.
Chronological 1: R&D (Practical) Perspective — Why This Architecture Is Practical
Architectural Changes (Details)Initial 3 stages based on convolution (rep mixer) rapidly compress local features.Subsequently, 2 stages of multi-head self-attention complement global relationships.An added downsampling stage (existing 4 stages → 5 stages) reduces input by 32 times to prevent token explosion.Operational AdvantagesThe structure itself, which reduces token count, drastically decreases TTFT, eliminating the need for ‘patchwork solutions’ like token pruning or tiling.Designed to process high-resolution images directly, reducing pre-processing overheads like tiling.Points Unmentioned by Other MediaApple optimizing the model for the device’s engine is not a simple conversion but a ‘system-level design’.Co-designing hardware (Neural Engine) and software (model architecture) dramatically accelerates its transition to real-world use.
Chronological 2: Industry & Market (Short-Term 6-12 Months) — Demand & Supply Changes
Hardware DemandPotential short-term slowdown in data center GPU demand.However, a surge in demand for edge NPU (Neural Processing Unit) and integrated SoC.Increased sales and upsell opportunities for high-performance laptops like MacBook Pro.Cloud Provider ImpactAs ultra-low-latency local models proliferate, some workloads will migrate away from the cloud.Cloud providers need to shift their revenue model to ‘on-device complementary services’ (model deployment, fine-tuning, synchronization).Software & Apple Ecosystem EffectsApp Store and service-based revenue will be strengthened (potential monetization of privacy-focused features).Expansion of developer tools and SDK market: Device optimization tools create new business opportunities.Points Often Missed by Other NewsGPU training demand won’t completely disappear.Large models and ultra-large-scale data training will still be in the cloud, but ‘deployment and real-world usage’ costs will dramatically decrease, reshaping capital flows.
Chronological 3: Economy & Investment (Mid-Term 1-2 Years) — Who Benefits
Benefiting IndustriesSemiconductors (edge NPUs, SoCs), hardware (high-performance laptops, AR devices), software (SDKs, apps), security & privacy solutions.Enterprise & Investor ChecklistIn semiconductors, focus on companies designing and producing edge NPUs instead of AI server GPUs.Cloud providers must have a strategy for transitioning to ‘hybrid services’.Software companies should secure a competitive edge with specialized services in model lightweighting and local fine-tuning.Labor Market ImpactIncreased automation pressure in high-frequency customer support and image-based inspection tasks.However, device-centric AI also has the potential to create jobs by fostering new apps, products, and services.Points Unmentioned by Other MediaThe spread of local AI implies a structural decline in ‘data transfer and storage costs’.This will accelerate AI adoption by SMEs, alongside service price competition.
Chronological 4: Structural Changes (Long-Term 3-5 Years) — Ecosystem, Regulation, National Strategy
Ecosystem RestructuringIf Apple commercializes a local-first model, a differentiated consumer experience based on privacy and speed will become the standard.Google, OpenAI, and other major players are likely to respond with similar features based on cloud or hardware partnerships.National Security & RegulationThe proliferation of on-device AI may ease issues related to cross-border transfer of personal information.However, the closed nature of local models (Apple’s exclusive distribution) could lead to competition restriction and standardization issues.Long-Term Investment StrategyFocus on edge NPUs, hardware acceleration libraries, and local privacy certification solutions.Standardization-related companies (model interoperability, format conversion tools) will also create new demand.Points Unmentioned by Other NewsFrom a policy perspective, new ‘local AI security standards’ will be needed.This means regulations on device internal model integrity and update systems are likely to emerge.
Risks and Counter Scenarios
If Apple Operates the Model in a Closed MannerTechnological advantages could lead to consumer lock-in, raising concerns about competition distortion.Potential hindrance to overall industry interoperability.True Costs of Training & TuningApple’s examples of small training nodes (e.g., 8×H180) are for ‘research’ costs, and large-scale commercial versions may require a different cost structure.Security & Verification IssuesLocal models could cause liability issues due to incorrect outputs (especially in OCR, medical, legal documents).SolutionsDevelopers must invest in model verification pipelines, update management, and signed model distribution systems.
Immediate Action Checklist for Enterprises & Investors (10 Items)
1) Product Team: Prioritize evaluating whether ‘local inference’ adds value to current products.2) Engineering: Secure personnel responsible for model lightweighting and Neural Engine optimization.3) Security Team: Establish model signing and integrity verification protocols.4) Sales Team: Design business models that convert privacy and low latency into a premium offering.5) Marketing: Prepare a ‘local AI-based privacy’ message.6) Investment Team: Re-evaluate portfolio of edge NPU, SoC, and AI SDK startups.7) HR: Plan for reallocation of repetitive back-office and OCR tasks.8) Policy Team: Draft compliance and regulatory risk scenarios related to model deployment.9) Partnerships: Accelerate discussions on collaboration with semiconductor and hardware manufacturers.10) R&D: Research Apple-style hybrid encoders and monitor related papers and patents.
Technical & Business Tips for Practitioners (Quickly Applicable)
Short-Term (0–3 Months)Experiment with scaling instead of tiling in high-resolution image workflows.Evaluate cost-performance by combining a small LLM (e.g., 2.5B class) with a FastVLM-like structure.Mid-Term (3–12 Months)Add local model execution options to product roadmap (customer-selectable).Establish on-premise and edge deployment testbeds.Long-Term (1+ Year)Formulate a strategy for proprietary lightweight vision encoders or consider acquiring startups specializing in such modules.Establish local AI certification and continuity management (OTA update system).
Economic Impact Summary Indicators (Projections) — Number-Based Scenarios
Short-Term (1 year): Potential for some cloud inference revenue (5–15%) to shift to edge/devices.Mid-Term (2 years): Edge NPU market annual growth rate (projected) accelerating to 20%+.Long-Term (3–5 years): Consumer ARPU rise due to increased high-value services based on software subscriptions and the App Store.Note: Figures vary significantly by case; company-specific sensitivity analysis is required.
Conclusion — Why This Event Is Not Just Technology News But Economic News
FastVLM is not simply a ‘faster model’.If local multimodal AI becomes a reality, cost structures, privacy norms, and product differentiation methods will change simultaneously.Enterprise strategies must be rapidly re-engineered from ‘cloud-centric’ to ‘hybrid/device-first’.Investors and policymakers must view the current situation as both an ‘opportunity’ and a ‘regulatory and standardization risk’ simultaneously.
< Summary >
Apple FastVLM ushers in the era of ‘local real-time multimodal AI’ by reducing TTFT by 85 times and shrinking the vision encoder by 3.4 times.The core lies in the Fast Vit HD architecture, which inherently solves token explosion, and its real-world usage demo on MacBook Pro.In the short term, there will be a rise in edge NPU and SoC demand, a partial shift of cloud inference, and a restructuring of software revenue.In the mid-to-long term, ecosystem lock-in, regulatory, and standardization issues will emerge, making the redesign of investment and corporate strategies essential.Immediate actions: Add ‘local AI options’ to product roadmaps, secure edge optimization personnel, and prepare security and update systems.
[Related Articles…]Summary of Apple’s AI Strategy Shift’s Impact on Korean SemiconductorsInvestment Strategy in the Era of Local AI: From Edge Chips to Software
*Source: [ AI Revolution ]
– Apple’s New AI SHOCKS The Industry With 85X More Speed (Beating Everyone)
● AI Agents’ Unique Identity Forces Governance Overhaul
Is AI Agent Identity Truly ‘Unique’? 6 Key Questions AI Agents Pose to Digital Workflows and Identity Governance
Key Takeaways from This Article Before You Read:
- This article proposes a method to distinguish between an AI agent’s unique identity and its instance.
- It provides practical solutions for legal, HR, and security risks arising when treating agents like employees.
- It specifically explains the pitfalls of using existing directories like Active Directory and offers alternatives (IAM, OIDC, service account patterns).
- It compares cost, governance, and logging strategies for persistent vs. ephemeral agent operating models.
- It identifies gaps in current IGA (Identity Governance and Administration) systems and presents a checklist for designing automation, policies, and verification needed in the era of mass agents.
- It includes proposals for ‘Agent Passport’, ‘provable provenance chain’, and ‘agent insurance/liability tiers’, which are rarely covered in other news or YouTube content.
1) Background — Differences Between Humans, Existing Non-Human Identities (NHI), and AI Agents
Humans are physical beings with affiliations, roles, and learning cycles within an organization.Existing Non-Human Identities (NHIs) are digital service accounts, automation scripts, API keys, etc., which are primarily deterministic and undergo few changes.AI agents are digital but can learn, make judgments, and design/modify their actions, thus having a human-like action cycle (evaluate → decompose → execute → learn).Therefore, ‘agent identity’ is not the same as NHI.SEO Keywords: AI Agent, Agent Identity, Non-Human Identity, Digital Workflow, Identity Governance.
2) Key Questions (in order) and Practical Interpretations
Sub-item — Question A: Is an agent just software?Main Content: Technically, it is ‘software’.However, behaviorally, it possesses decision-making, autonomy, and a learning cycle, so treating it merely as a service account will lead to failures in control, accountability, and auditability.Practical Proposal: Model agents as ‘service identity + behavioral metadata’.Examples of Essential Metadata: owner, purpose, scope, modelversion, memorylevel, TTL (Time-To-Live).
Sub-item — Question B: Should agents be recognized as coworkers?Main Content: Absolutely not in a legal employment relationship sense.However, treating them as ‘virtual coworkers’ by assigning roles within teams for specific tasks and leaving entries in collaboration tools (e.g., chat logs, task assignments) can be useful.Caution: HR benefits, insurance, and employment law application must be prohibited, and clear disclosure (“This result was generated by an agent”) must be ensured system-wide.
Sub-item — Question C: Should agents be put into directories? (AD, LDAP, etc.)Main Content: Indiscriminately adding them to traditional directories increases management and audit burdens, as well as security risks.Practical Proposal: Use a hybrid approach of IAM (cloud IAM, OIDC, service accounts) + a dedicated agent registry (metadata DB) rather than physical AD.Example: For long-term/persistent agents, use IAM service accounts; for one-time/mass spin-up agents, use OIDC-based temporary tokens/credentials.
Sub-item — Question D: Should agents be persistent or ephemeral?Main Content: The answer is ‘a mix depending on the use case’.
- Persistence needed for: customer-facing chatbots, agents requiring long-term context retention.
- Ephemerality recommended for: data analysis, batch jobs, single-task automation.Recommendation from a cost, security, and governance perspective: Default to an ‘ephemeral-first’ policy, allowing long-term instances only after procedural approval, budget allocation, and SLA agreement.
Sub-item — Question E: Is the current IGA system sufficient?Main Content: Most organizations’ IGA systems are ‘human-centric’ in design, making them vulnerable for managing mass, temporary, and dynamic agents.Essential Improvement Areas: Automated provisioning/deprovisioning, automated annual/periodic attestation for each agent, entitlement request/approval processes, log/provenance retention.Technology Stack Proposal: Policy-as-code (OPA), event-driven approval workflows, console-based audit dashboards.
3) The Most Important Points Rarely Covered in Other News/Videos
Need to introduce the concept of an Agent “Passport”.Main Content: Each agent instance must have a ‘passport’.Passport Items: Agent ID, model/data source, training date, risk level, approver, TTL, cryptographic signature.Effect: Ensures auditability, accountability, and reproducibility; provides evidence for post-incident analysis.
Need ‘monetary and legal liability lines’ for agent decisions.Main Content: Insurance/budget allocation and liability classification (operator, owning team, vendor) must be pre-defined to prepare for damages caused by AI’s misjudgment.
Provenance chain and result watermarking are essential.Main Content: Results generated by agents must automatically include metadata indicating ‘which model and data generated them’.Reason: Regulatory compliance (e.g., finance, healthcare), building customer trust, enabling rollback in case of erroneous decisions.
New scenarios for agent hijacking/spoofing risks.Main Content: If agent authentication information (tokens) is stolen, accounts acting ‘like humans’ can be exploited.Response: Short TTL, behavior-based anomaly detection, policy guardrails for actions.
4) Technical & Security Practice Checklist (Immediately Applicable → 3 Months → Long-term)
Immediate (0–1 month)
- All agents must register at least owner, purpose, and TTL as metadata.
- Centralize (immutable) agent activity logs.
- Secrets management must only be allowed through a dedicated secrets manager.
Short-term (1–3 months)
- Establish a dedicated approval workflow for agents (automated).
- Implement least privilege and ABAC policies for each agent.
- Design an agent passport schema and introduce a signing mechanism.
Mid-term (3–6 months)
- Automate mass provisioning/attestation integrated with IGA.
- Calculate performance and risk scores and implement dashboards.
- Standardize provenance metadata for agent-generated outputs.
Long-term (6–12 months)
- Manage agent lifecycle as policy-as-code.
- Establish compliance processes linked to external regulatory/legal requirements.
- Consider third-party certification for supply chain (model, data) security and verification.
5) Operating Model Proposal — Agent Classification and ID Policy
Classification (Simple Version)
- Type A: Persistent Context Agent (e.g., customer support, requires permanent memory).
- Type B: Task-based Ephemeral Agent (analysis, batch processing).
- Type C: External Vendor/Partner Agent (most challenging to manage and audit).
ID Policy (Recommended)
- Agent instances should be created with a unique ID (UUID) + organizational code.
- Create type-specific permission templates to standardize provisioning.
- Prioritize the use of OIDC/PKI-based temporary tokens for all authentication.
6) Cost and Governance Trade-offs
Persistent agents offer advantages like fast response and context retention but incur continuous infrastructure costs (CPU, memory, storage) and increased security costs.Ephemeral agents are resource-efficient but involve context re-initialization overhead and may experience cold start delays.From a governance perspective, a large number of ephemeral instances can explosively increase IGA and audit burdens, so policies for provisioning limits, quotas, and automatic garbage collection must be established.
7) Recommendations from a Legal, Regulatory, and Organizational Policy Perspective
Legal: Ensure agent-generated decisions are under the ‘human oversight’ principle.HR: Do not classify agents as employees; manage them as ‘system assets’.Contracts: Include model version, data provenance, and liability sharing clauses in vendor contracts.Insurance: Consider ‘AI risk insurance’ for agents involved in critical business decisions.
8) Examples by Operating Scenario (Quick Application Tips)
Customer Support Automation
- Persistent context retention is advantageous for agents.
- However, when accessing sensitive information, ‘temporary permission granting’ + human handover (workflow) is essential.
Financial Report Automation Assistant
- Agent passport and provenance chain are mandatory.
- All outputs must be recorded as signed records to comply with regulatory audits.
Data Analysis Batch Processing
- Ephemeral agents + automatic decommissioning recommended.
- Set spot instance and quota configurations for cost optimization.
< Summary >
AI agents are technically software, but due to their behavioral and governance characteristics, they require a separate identity model distinct from existing non-human identities.While agents should not be treated as employees, allow for ‘virtual coworker’ level collaboration aspects to clarify collaboration, audit, and accountability.Do not indiscriminately put them in directories; adopt a hybrid model based on IAM, service accounts, and OIDC.Determine persistence on a case-by-case basis, but an ‘ephemeral-first’ policy is generally recommended.Current IGA systems need enhancement; introduce agent passports, provenance chains, policy-as-code, and automated verification processes.Furthermore, establishing an ‘Agent Passport’ and ‘legal and monetary liability lines’—points rarely discussed elsewhere—will significantly enhance an organization’s competitiveness and regulatory compliance.
[Related Articles…]
-
AI Agent Identity: How Should Companies Prepare?AI Agent Identity: How Should Companies Prepare?
-
Identity Governance and Security Strategy in the Agent EraIdentity Governance and Security Strategy in the Agent Era
*Source: [ IBM Technology ]
– Are AI Agent Identities Really Unique? AI’s Role in Digital Workflows
● Greenblatt’s Magic Formula- Cheap Quality, Historic Goldmine
Why are Good Companies Underpriced? Now is a Historic Buying Opportunity — Joel Greenblatt’s Magic Formula and Practical Strategies
This article systematically covers the following key contents.
The principles of the Magic Formula (composition, scoring, rebalancing) and actual backtest results.
The real reason why the Magic Formula works — the alpha generated by ‘practical feasibility’ and investor patience, not just statistics.
Structural differences between institutions and individuals, and a realistic explanation of why individuals have an advantage in long-term investing.
Reinterpretation of the identity of growth vs. value and the margin of safety (a vague but practical application method).
Reasons for the widening value spread in the current market and practical portfolio design methods to convert this into a buying opportunity.
Crucial details often not covered by other news or YouTube — the pitfalls of long/short (beta imbalance), specific training methods to cultivate patience, and the core reasons why individual investors fail in practice, all included.
1) Magic Formula Summary: What to Look for and Why
Simply put, the Magic Formula is a system that identifies ‘companies that earn good money and are currently undervalued (undervalued stocks).’It combines two key metrics.First, Return on Invested Capital (ROIC, etc.) — how much profit a company generates relative to the capital it has invested.Second, Earnings Yield — the ratio of earnings to enterprise value, representing the relative size of earnings compared to the stock price.The basic rule is to rank these two metrics separately, then sum the ranks and buy the top 20-30% with the highest scores.Rebalancing is typically done once a year.This method falls under value investing (SEO keyword) but is executed mechanically like a quant strategy.
2) Backtesting and Real-World Performance: What the Numbers Say
In Greenblatt’s published backtest from 1988–2004, the Magic Formula portfolio significantly outperformed the S&P 500.Even when including extended research periods (up to recent times), results still report outperformance of market returns.It also showed performance exceeding expected returns in terms of CAPM alpha.But an important fact: performance appears in the long term, and short-term periods of underperformance (3 years or more) can exist.
3) Why a Simple Formula Generates Excess Returns — Key Reasons
It’s because not everyone can put into practice a formula that everyone knows.Even under the efficient market hypothesis, ‘knowing but not doing’ exists.Key factors: psychological factors (lack of patience), institutional constraints (short-termism of institutions), and execution failure (timing pursuit).Ultimately, excess returns do not come from the secrecy of the algorithm itself, but because few people put it into practice even when they know it.In other words, ‘practical feasibility’ is the momentum of this strategy.
4) The Power and Reality of Patience (Long-Term Investing) — Structural Advantages Held by Individuals
Institutions structurally find it difficult to sustain long-term investing.Many institutions have an average holding period of less than one year, and there are cases where even funds and pension funds cannot endure three years of underperformance.Individuals have no mandatory redemptions and can maintain long-term positions even with small amounts, giving them a relative advantage.However, patience is not as easy as it sounds.The reason why investors’ average investment returns were low despite famous funds having recorded high average annual returns in the past was due to ‘buying high and selling low’ (chasing rallies and panic selling).Therefore, the core competitiveness in long-term investing (SEO keyword) is ‘the ability to endure.’
5) Practical Tips for Cultivating Patience — Emotion Management and the Role of Knowledge
Knowledge reduces emotional volatility.Documenting your investment logic (data, assumptions, ranges) provides a standard when you waver.Specific methods: drafting investment memos (assumptions, discount rates, worst-case scenarios), establishing a 3-year plan, setting annual review rules.Furthermore, by keeping records of profits and losses and repeatedly reviewing past instances of ‘buying at a low point → being rewarded,’ psychological resilience is strengthened.
6) Know Thyself — Three Elements for Successful Individual Stock Investing
Joel states that three things are necessary for individual stock investing.Accurate valuation ability.The ability to set a sufficient Margin of Safety.’Execution power (patience and discipline)’ to carry this out over the long term.If you don’t possess these three, you should opt for an index (passive) approach.In other words, understanding your scope of competence is the first step.
7) Reinterpretation of Growth Stocks vs. Value Stocks — The Essence is the Same
Growth stocks and value stocks are essentially on the same spectrum.The difference lies only in the time horizon and expected growth rates.The margin of safety is not an absolute figure but varies depending on ‘the outlook for future growth (or decline).’Therefore, in value investing, the margin of safety should be defined as ‘the difference between the current price and my projected future value.’
8) Practical Application: Portfolio Design and Management Rules
Screening: Score by combining top ROIC + top Earnings Yield across the entire market.Purchase: Include the top 20-30% of companies with equal-weighting.Allocation: For individual accounts, diversification into 20-40 stocks is recommended.Rebalancing: Once a year.Cash Management: Reserve some cash during extreme market booms.Avoid shorts — The Pitfalls of Long/Short: Due to differences in beta and volatility between long and short positions, a de facto leverage effect occurs, making risk opaque.Individuals are generally advised to only go long.
9) Real Risks and Practical Limitations of the Magic Formula
Possibility of long drawdowns (3 years or more).ROIC may continuously deteriorate due to structural changes by sector/country (e.g., fundamental changes in industries).Increased transaction costs due to small cap companies and illiquidity issues.Excessive backtest fitting (designing parameters that only fit the past) leads to errors.Therefore, pre-assumptions and stress tests for risk management are essential.
10) Why Now is a Historic Buying Opportunity — Value Spread and Market Environment
In the current market, the value spread between ‘the most expensive stocks’ and ‘the cheapest stocks’ has significantly widened.This suggests that concentration in certain large-cap growth stocks and the neglect of many other companies are occurring simultaneously.Historically, such spreads have shown mean reversion, and a top-score-based buying strategy (Magic Formula) has generated profits.Important: This opportunity is likely to be effective with a strategy of ‘immediately selecting the high-scoring group and holding it with annual rebalancing.’
11) Ready-to-Use Checklist (Actionable)
1) Screener Setup: Score based on ROIC and Earnings Yield criteria.2) Document Portfolio Rules: Formalize buy, rebalance, and exit rules in writing.3) Psychological Rules: Establish a rule for allowing 3-year drawdowns (e.g., review procedure if account value drops by 30%).4) Exposure Management: Diversify by industry and market capitalization.5) Reporting and Review: Annual qualitative re-evaluation (business model changes, ROIC changes, major risks).
12) Crucial Points Rarely Covered in News — Where the Game is Won or Lost
The most important fact is not ‘the strategy itself’ but ‘whether one can execute the strategy to the end.’Many YouTube videos and news articles only show backtest performance, but actual alpha goes to ‘the few who endure periods that the majority cannot.’Furthermore, the fatal danger hidden behind the intuitive appeal of long/short strategies (de facto leverage due to beta mismatch) is often overlooked.Finally, when setting a margin of safety, practical sophistication is required to consider ‘future value changes’ and ‘the range of uncertainty’ together.
< Summary >The Magic Formula is a simple yet powerful value investing strategy that mechanically selects undervalued stocks by combining ROIC and Earnings Yield.The key advantage is not the algorithm but the patience of investors who survive and put it into practice.Due to institutional structures making long-term investing difficult, individuals now have an advantage through long-term investing (patience).Growth and value are merely issues of time horizon, and the margin of safety must account for future value changes.The current market’s value spread is likely to present a historic buying opportunity, so a calm approach with an annual rebalancing-based long-only portfolio is a realistic strategy.
[Related Articles…]Value Investing: Is Now the Time? — A Summary of Buying Timing Based on Value SpreadsLong-Term Investing and Patience: The Paradox of Quant Strategy — Practical Rules for Individuals
*Source: [ 에릭의 거장연구소 ]
– 왜 좋은 기업이 헐값일까? 지금이 역사적 매수 기회다. (조엘 그린블라트)
Leave a Reply