Week in Tech #8 – August 18, 2025
AI Bubble Chatter, EGI over AGI, and Agentic Patterns Hitting the Enterprise
This week main focus was on: Valuations vs. value in AI (bubble talk), practical “Enterprise General Intelligence” (EGI) vs. AGI hype, and the maturation of agentic patterns (MCP, Agent Factory) across cloud and DevOps stacks—plus steady moves on language data, observability, and European cloud sovereignty.
Executive Summary
Walk into Monday’s boardroom and the word “bubble” hangs in the air. Yet when the spreadsheets open, budgets still flow to what returns cash today: agentic workflows stitched into core systems, network effects around data, and domain‑tuned models that move real KPIs. It’s ROI over rhetoric.
Down on the operations floor, the conversation changes. AGI slides give way to Enterprise General Intelligence—composable agents with guardrails, governance, and measurable outcomes woven through CRM, ERP, and service desks.
In the engine room of the stack, the pieces click into place. Microsoft’s Agent Factory lays out patterns; MCP moves from idea to instrumentation—Sentry is monitoring servers, HashiCorp explores Terraform integrations—and Azure ties the whole thing into observability. For the first time, an enterprise reference architecture for agents feels real.
In the data archives, the moat deepens. NVIDIA ships a high‑quality European‑language corpus and tools, a reminder that curated, regional data—not just ever‑larger models—wins accuracy and trust.
Meanwhile, in Brussels, the plot thickens. Cloud sovereignty debates force leaders to balance residency and compliance with vendor leverage and global scale.
And on the test track, the model race has no single champion. Early GPT‑5 vs. GPT‑4o head‑to‑heads trade wins. The craft returns to basics: choose per workload—reasoning depth, latency, and cost—not by version number.
Future Predictions
1. Agent observability becomes table stakes
MCP-aware monitoring (errors, tool-call latencies, hallucination rates, safety events) will be built into APM and SIEM by default.
2. Pattern libraries, not proofs-of-concept
“Agent Factory” style blueprints (retrieval, orchestration, human-in-the-loop, safety gates) become standardized platform features in Azure/AWS/GCP and dev platforms.
3. Data-native localization
Investment shifts from generic frontier models to region/language-specific data pipelines (EU languages, sector ontologies) to win accuracy and trust.
4. Sovereign-by-design architectures
Multi-cloud patterns will include residency controls, in‑region inference/endpoints, and portable IaC contracts to satisfy EU sovereignty without stalling velocity.
Resources
Sam Altman on an AI Bubble
A candid take that current AI valuations may outrun near‑term value capture—useful context for budgeting and ROI discipline.
Read more
Is GPT‑5 really worse than GPT‑4o? (Ars Test)
Early, mixed benchmarking across tasks; reinforces a workload‑first approach to model selection (cost, latency, reasoning).
Read more
Humanoid Bots & Manufacturing Disruption
NVIDIA, Foxconn, and Tesla push factory‑scale humanoid robotics—pointing to automation stepping beyond arms and AMRs.
Read more
AGI Debate vs. Enterprise General Intelligence (EGI)
Argument that focusing on EGI—agents integrated with business systems—delivers outsized near‑term value vs. abstract AGI.
Read more
NVIDIA’s EU Language Dataset & Tools
A large, high‑quality European‑language corpus and tooling; fuel for localized accuracy, safety, and trust.
Read more
Sentry adds MCP Server Monitoring
Observability for MCP servers lands in a mainstream APM—signals rising operational maturity for agentic stacks.
Read more
HashiCorp & MCP Servers for Terraform
HashiCorp explores MCP‑aligned servers/plug‑ins to wire agents with IaC workflows—IaC meets agent orchestration.
Read more
Azure “Agent Factory” Patterns
Microsoft documents common agentic use cases and design patterns (retrieval, tool orchestration, oversight) for enterprise delivery.
Read more
European Cloud Sovereignty
Round‑up of EU sovereignty moves and what they mean for workload placement, data governance, and vendor strategy.
Read more
Stay curious, Maciej Gos