Albin George
AG Radar
Anyone can claim foresight after the outcome. This page removes that option. Each blip is a market call, made before the result and kept on screen whether it lands or misses. Signals is where I write. AG Radar is where I keep score. All views are personal.
One falsifiable sentence with a deadline. On the radar the day it is made, never edited after.
Every quarter, each call is checked against the evidence for it and against it.
Calls resolve as playing out, early, or wrong. The wrong ones stay on screen. That is what makes the rest worth reading.
Distance from centre is the resolution date. Hover a blip for the call, tap for the full reasoning.
Every quarterly sweep updates each ring's colour and adds a dated review note inside the call, so the verdict trail lives right here on this page. Each call's runway progress appears when you hover on or open it. First sweep lands October 2026.
By December 2028, at least one Big Four network will publicly market a named subscription or platform offer for recurring tax, risk or finance work in which AI agents are described as the standard delivery workflow and human experts are positioned primarily as reviewers, exception handlers or escalation points.
Rules-heavy and recurring work is where agentic delivery should become commercially visible first. The Big Four are already building agent-orchestration layers inside their delivery platforms, but most client packaging still reflects projects, teams and managed-service capacity. I believe the operating model will change before the firms are fully comfortable admitting that the human pyramid is no longer the product.
I would abandon the call if the firms continue using agents only as internal productivity tools or bespoke implementation components, without changing how recurring work is packaged for clients. Persistent regulatory liability, weak agent reliability or client insistence on named human delivery teams would also undermine the thesis.
By the publication of fiscal-year results covering periods ending no later than December 2028, at least one global consulting or professional-services company with more than US$10 billion in annual revenue will report two consecutive fiscal years of revenue growth alongside flat or lower year-end headcount, and management will explicitly attribute part of the resulting productivity or operating leverage to AI-enabled delivery.
Professional-services economics have historically linked revenue growth to adding people. AI does not need to eliminate jobs for that relationship to weaken; it only needs to let each team support more revenue, reuse more intellectual property and automate enough production work that hiring no longer tracks demand proportionately. I expect that change to become visible in reported economics before firms describe it as a permanent redesign of their workforce.
I would reconsider the call if meaningful revenue growth continues to require proportional workforce expansion, or if companies achieve flat headcount only through cyclical layoffs while revenue stagnates. I would also treat the call as unproven if management attributes the leverage solely to pricing, geographic mix, utilization or conventional offshore delivery rather than AI-enabled productivity.
By December 2027, at least one of Gartner, Forrester, IDC, Omdia or Everest Group will position a conversational, AI-native interface to its licensed intellectual property as the lead paid-access product in its enterprise commercial packaging, with reports and analyst enquiries presented as supporting access modes rather than the primary product.
Research buyers increasingly need answers assembled around a live decision rather than another document to read. The defensible asset is the provider's evidence base, taxonomy and accumulated judgment, not the PDF container. Once clients become accustomed to interrogating enterprise knowledge conversationally, research firms will have to sell direct access to their intelligence rather than lead with a library of publications.
I would abandon the call if conversational access remains an optional search feature inside conventional subscriptions and the commercial hierarchy continues to lead with reports, seats and analyst calls. Poor answer reliability, weak attribution or client reluctance to trust generated synthesis over source material would also weaken the thesis.
By December 2027, at least one of Gartner, Forrester, IDC, Omdia or Everest Group will commercially launch a secure MCP server, enterprise connector or equivalent governed API that allows customer-controlled copilots or AI agents to retrieve and use the provider's licensed proprietary research directly inside enterprise workflows.
The centre of gravity in knowledge work is moving away from individual vendor portals and toward the enterprise's own copilot or agent layer. Research providers may prefer to control the interface, but clients will increasingly expect licensed intelligence to appear where decisions are being prepared. Distribution into governed enterprise agents should therefore become as strategically important as access through the provider's website.
I would abandon the call if licensing, attribution and leakage risks cause the major research firms to keep their content inside proprietary interfaces. The call would also weaken if enterprises decide to connect their agents directly to raw company filings, news and internal data, reducing the value of third-party research connectors.
By December 2028, at least one engineering and R&D services provider with more than US$500 million in annual revenue will publicly make outcome-linked commercial terms a standard contracting option within a named AI-led engineering offer, with fees tied to measurable engineering KPIs such as development cycle time, test coverage, uptime, defect reduction or certification milestones.
AI-led engineering claims will eventually have to move beyond demonstrations of productivity and into commercial accountability. Engineering workflows generate measurable outputs, and providers that genuinely believe their platforms shorten development or improve asset performance should be able to place part of their fees at risk against those results. Doing so would also help providers defend value as effort-based pricing comes under pressure.
I would abandon the call if customers continue to insist that product architecture, supplier delays and internal decision-making make engineering outcomes too difficult to attribute to one provider. Liability, certification risk and clients' reluctance to share operational data could also keep commercial models anchored in time-and-materials, fixed scope or capacity-based contracts.
By September 2027, at least one hyperscaler will publicly fund, co-invest in or reserve for multiple years a dedicated production line or defined tranche of manufacturing capacity directly with a transformer or medium- or high-voltage switchgear manufacturer.
Power equipment is shifting from an ordinary procurement category to a constraint on the pace at which computing capacity can be deployed. When a component's lead time can determine when billions of dollars of data-centre infrastructure becomes productive, the largest buyers have an incentive to secure manufacturing capacity rather than rely only on project-by-project orders placed through utilities, developers or contractors.
I would abandon the call if utilities and engineering contractors retain exclusive control over grid-equipment procurement despite hyperscaler pressure. Faster manufacturing expansion, standardized modular equipment or a decisive move toward off-grid generation could also reduce the need for hyperscalers to reserve OEM capacity directly.
By June 2028, at least three G20 governments other than the United States and United Kingdom will adopt a statutory, regulatory or cabinet-approved grid-connection or strategic power-equipment manufacturing measure that explicitly cites AI or datacentre load growth as a reason for the intervention.
Generic grid reform becomes politically difficult once scarce capacity has to be allocated among households, conventional industry, electrification and large computing loads. Governments that view domestic AI infrastructure as strategically important will eventually have to name the demand they are trying to accommodate, whether through priority connection rules, flexible-load regimes or support for transformer and switchgear manufacturing.
I would abandon the call if governments consistently keep reforms technology-neutral and refuse to give AI infrastructure identifiable treatment. Strong public resistance to data centres, electricity-price concerns or a shift toward privately powered and off-grid facilities could also prevent national grid policy from naming AI load directly.
By September 2027, at least two additional mass-market OEM groups headquartered outside China, excluding Hyundai Motor Group and Stellantis, will have launched, homologated or formally announced with a named launch market and date an extended-range electric vehicle for sale in North America, Europe, India, Japan, South Korea or Australia.
The transition to full battery-electric vehicles is developing unevenly across markets. EREVs give established manufacturers a way to offer electric driving for most journeys without depending on uniformly mature charging networks or very large batteries. I expect more global OEMs to treat the architecture as a pragmatic bridge rather than a China-specific product category.
I would abandon the call if battery costs fall and charging availability improves quickly enough to make the additional engine and fuel system commercially unnecessary. Regulatory treatment that gives EREVs little advantage over conventional plug-in hybrids, or poor consumer acceptance of their complexity, would also weaken the case.
Each signal is one falsifiable statement, dated when it was made and never edited after the fact. Statuses are reviewed quarterly: Playing out means evidence is confirming the call, Early means the direction looks right but the timeline was optimistic, Watching means too soon to judge, Wrong means the call missed and stays on the board.
Calls draw only on public data and broad industry currents. Nothing here concerns specific securities or constitutes investment advice. Where a call touches the market my employer operates in, the employer is excluded from scoring, and no call is evidenced or resolved using employer material. All views are personal.