top of page

Week in IT Digest #69


IT digest summary

The single most important news this week: the industry coalescing around open standards for AI agents (AAIF with MCP and peers). Why it matters: interoperability makes agents practical in production—cutting integration time, reducing lock-in, and unlocking safer, governed automations that entrepreneurs and developers can ship faster at lower cost.

TL;DR

  • Agentic AI is standardizing fast (AAIF + Google’s managed MCP), making cross-service integrations plug-and-play.

  • Model dominance is fragmenting: open-weights (Mistral) rise, Meta hints at going closed, Anthropic gains enterprise share.

  • Security risk escalates: AI-assisted social engineering, unfixed legacy vulns, and stronger offensive model capabilities.

  • Compute supply is geopolitically constrained: partial H200 access to China, smuggling crackdowns, RAM shortages driving costs up.

  • Enterprise ROI is real: Adobe’s AI features lift revenue; investors frame AI as long-term economic infrastructure.

  • Prepare for agent ops: identity, policy, observability, and licensing/ethics checks will join your deployment pipeline.

Change Summary

The center of gravity is moving from raw model supremacy to platform orchestration: standards (AAIF, MCP) and integrations (Adobe-in-ChatGPT) reduce integration friction and vendor lock-in, shifting value to those who can compose agents with identity, governance, and runtime context. As interoperability hardens, switching costs fall and multi-model strategies become tractable—raising competitive pressure on single-vendor stacks while expanding the total addressable market for agent-native apps.

Simultaneously, compute access and security now define execution risk. Policy whiplash (partial H200 access to China, tariffs, smuggling busts) collides with swelling demand (RAM/GPU shortages), pushing budgets up and timelines out. Organizations that abstract model choice, plan for burstable/heterogeneous compute, and bake in AI-aware security controls will run faster and safer. Expect procurement to converge with AppSec: model/agent SBOMs, ethics certifications (IEEE), and content-licensing signals emerge as checklists alongside SOC2 and ISO—forcing product and legal to collaborate earlier in the build cycle.

Second-order effects: developer tools will look more like ops stacks. Agent identity, policy, and observability will become the “Kubernetes” of AI, enabling safe autonomy through cryptographic IDs, guardrails, and runtime telemetry (e.g., AgentField, Lightrun). As enterprises see AI lift revenue (Adobe) and reports position AI as economic infrastructure (PitchBook), the default roadmap tilts toward embedded agents across workflows. Teams that treat agents as products—not prompts—will compound advantages through data flywheels, safety feedback loops, and cross-service automations.

Change Patterns

The prior 10 weekly digests are unavailable (null), so trend analysis relies on current week signals and longer-running industry context. What stands out this week versus recent months: (1) a concrete pivot from hype to standards—AAIF and Google’s managed MCP move agent interoperability from slideware to shippable; (2) vendor power is rebalancing—Anthropic gains enterprise share, Mistral’s open-weights coding model closes on closed alternatives, and Meta weighs a proprietary turn; (3) security incidents keep rising in parallel with capability advances—zero-days in core dev infra, AI-assisted malware delivery, and warnings about growing cyber potency of frontier models; (4) compute scarcity remains structural—policy shifts (H200 to approved Chinese buyers) and enforcement (smuggling arrests) coexist with component shortages that spill over into consumer and enterprise hardware; (5) enterprise monetization is material—Adobe’s earnings and sustained mega-investments (Microsoft/Amazon in India) point to durable AI capex and revenue momentum. Patterns that persist: regulatory and licensing pressure intensifies (EU probes, ethics certifications, publisher licensing specs), and the open vs. closed oscillation continues as firms optimize for speed, safety, and margin. Net: the trendline is toward interoperable agents, governed by emerging standards and compliance layers, deployed atop scarce, geopolitically sensitive compute—rewarding teams that design for portability, security, and cost from day one.

Topic Clusters

Agents become a platform: standardization and integrations

  1. OpenAI, Anthropic, and Block join new Linux Foundation effort to standardize the AI agent era read full article

  2. Major players are contributing MCP, Goose, and AGENTS.md to the Agentic AI Foundation (AAIF), aiming to make agent frameworks interoperable and reduce fragmentation.

  3. Google launches managed MCP servers that let AI agents simply plug into its tools read full article

  4. Google released fully managed, remote MCP servers so agents can natively connect to Maps, BigQuery, Compute Engine, and GKE—accelerating real-world agent deployments.

  5. Adobe brings Photoshop, Acrobat and Adobe Express to ChatGPT read full article

  6. ChatGPT now hosts Photoshop, Acrobat, and Express via MCP-powered apps, enabling context-aware creative and document workflows inside the chatbot.

  7. AgentField aims to fix agentic AI’s coordination crisis with cryptographic IDs and Kubernetes-style orchestration read full article

  8. AgentField launched an open-source stack for agent identity and orchestration, targeting scalable, governed agent operations in production.

The model race recalibrates: open-weights rise, incumbents pivot

  1. OpenAI's house of cards seems primed to collapse read full article

  2. OpenAI’s dominance is challenged by Anthropic, Google’s Gemini 3 Pro, and China’s DeepSeek, with GPT-5 underwhelming and pressure mounting to regain leadership.

  3. Mistral AI’s Devstral 2 is an open-weights vibe coding model built to rival the best proprietary systems read full article

  4. Devstral 2 (123B params) targets autonomous software engineering with open weights, closing the gap with top closed models.

  5. Meta is reportedly working on a new AI model called 'Avocado' and it might not be open source read full article

  6. Meta may shift away from open-source with 'Avocado' (expected 2026), signaling a strategic pivot amid Llama pains and intensifying competition.

  7. Menlo Ventures: business spending on generative AI hit $37B in 2025, up from $11.5B in 2024; Anthropic's share grew from 24% to 40% YoY read full article

  8. Enterprise genAI spend more than tripled YoY, with Anthropic rapidly gaining enterprise LLM share—evidence of shifting vendor dynamics.

Security shockwaves: AI-enabled attacks and unresolved vulns

  1. Hackers tricked ChatGPT, Grok and Google into helping them install malware read full article

  2. Attackers seeded harmful terminal commands via AI-generated Q&A to social/search, exploiting user trust to spread malware.

  3. 700+ self-hosted Gits battered in 0-day attacks with no fix imminent read full article

  4. A zero-day is actively compromising hundreds of self-hosted Git instances with no immediate patch, underscoring supply-chain risk.

  5. Microsoft won’t fix .NET RCE bug affecting slew of enterprise apps, researchers say read full article

  6. Researchers warned of RCE via legacy .NET proxy behavior; Microsoft reportedly declined a fix, leaving long-tail enterprise exposure.

  7. OpenAI says the cyber capabilities of its frontier AI models are accelerating and warns that upcoming models are likely to pose a 'high' risk read full article

  8. OpenAI cautions that newer models’ cyber power is rapidly increasing—elevating offensive risk profiles.

Policy, ethics, and platform governance

  1. Two New AI Ethics Certifications Available from IEEE read full article

  2. IEEE launched CertifAIEd for professionals and products, mapping to legal regimes (e.g., EU AI Act) across accountability, privacy, and bias.

  3. EU opens antitrust investigation into Google's AI practices read full article

  4. Brussels is probing Google’s use of web/YouTube content for AI without fair compensation or opt-outs and potential impacts on rivals.

  5. Really Simple Licensing spec lets web publishers demand their due from AI scrapers read full article

  6. A new spec aims to let publishers signal licensing terms to AI scrapers, pushing toward remuneration and consent models.

  7. Tim Cook lobbies against app-store age verification mandates read full article

  8. Apple argues age-verification laws risk privacy by necessitating sensitive data collection; similar laws are spreading globally.

Compute, chips, and geopolitics

  1. NVIDIA can now sell its high-end AI chips to 'approved customers in China,' Trump says read full article

  2. US allows Nvidia’s H200 sales to approved Chinese buyers with a 25% tariff, keeping top-end chips restricted—reshaping supply calculus.

  3. Texas authorities have made multiple arrests in an NVIDIA GPU smuggling operation read full article

  4. A $50M GPU smuggling ring to China was disrupted, highlighting the gray market and export-control frictions around AI compute.

  5. Your Next Laptop Could Cost More. The RAM Shortage Is to Blame read full article

  6. OEMs warn of price hikes due to memory shortages as data center AI demand soaks supply, pressuring consumer and enterprise hardware budgets.

  7. Sources: ByteDance and Alibaba have asked Nvidia about placing large orders for H200 chips read full article

  8. China’s tech giants are moving to bulk-order H200s post-export approval, racing to secure constrained compute.

Enterprise AI adoption and investment momentum

  1. Adobe beats earnings expectations as AI tools drive double-digit revenue growth read full article

  2. AI-enhanced offerings lifted Adobe’s Q4 revenue 10% YoY with strong outlook—proof that AI features are converting to revenue.

  3. PitchBook: AI is becoming the defining infrastructure layer of the global economy read full article

  4. Long-horizon thesis: AI underpins enterprise software, security, defense, and supply chains over the next five decades.

  5. Microsoft to spend $17.5B on AI infrastructure to support India’s AI ambitions read full article

  6. Microsoft is investing heavily in India’s AI/data center capacity, signaling strategic regional expansion.

  7. Amazon to invest $35B in India by 2030 with focus on AI, logistics read full article

  8. Amazon’s long-term $35B bet emphasizes AI-driven logistics and cloud growth across the Indian market.

bottom of page