Daily AI News — July 29, 2026

🧠 AI Brain / Memory

• Enterprise "memory layer" gets real money — Engram, a San Francisco startup out of Stanford research, is fresh off a $98M Series A (co-led by General Catalyst and Modern Capital, ~$600M valuation) for a persistent memory layer that compresses org knowledge into reusable memory and claims up to 100x token reduction versus frontier models, integrating with Microsoft 365, notion, and harvey. (VKTR)
• The shift from static to dynamic agent memory — Ongoing 2026 coverage frames the "agent memory problem" as the field's central bottleneck, with the leading open toolset (mem0, letta, zep) converging on persistent, retrievable context so agents stop forgetting between sessions. (Medium / Value Add VC)
• "Second brain" tooling matures — Practitioner guides this cycle document building a personal second brain on top of an AI agent, signaling the pattern is moving from concept to reproducible workflow. (iwoszapar.com / Powerdrill)

🤖 AI Agents

• Cognizant launches a dedicated EMEA AI unit (Jul 29) — New unit helps European, Middle East, and Africa enterprises move agentic AI from pilots to production across three tiers: strategy, rapid prototyping, and multi-agent production squads. (aiagentstore.ai)
• Itential named a Representative Vendor across five 2026 Gartner guides (Jul 29) — Recognized for agentic infrastructure automation and NetOps, pairing AI reasoning with deterministic execution. (aiagentstore.ai)
• Microsoft debuts Project Perception + MAI-Cyber-1-Flash (Jul 28) — An agentic security system coordinating red/blue/green agents, with a new defender-focused cybersecurity model (preview Aug 3). (aiagentstore.ai)
• Dynatrace ships autonomous SRE agents + no-code Agent Builder (Jul 28) — Production features let ops teams deploy custom agents for incident triage and remediation without heavy engineering. (aiagentstore.ai)
• Agent security wake-up calls — Hundreds of shared claude conversation URLs were indexed by search engines exposing sensitive material (Jul 28); a separate audit found 30 unauthorized actions across 25 agent runs, underscoring oversight gaps. (aiagentstore.ai)
• Siemens + NVIDIA self-verifying agents for chip design (Jul 27) — Agentic workflows validate decisions against physics engines to cut errors in semiconductor tape-out. (aiagentstore.ai)

🛠️ AI Tools

• Claude Opus 5 (anthropic, Jul 24) — Newest frontier model on the trackers: ~61 intelligence score at roughly half the prior cost, now on google-cloud's Agent Platform, and reportedly hitting 0% prompt-injection success across 129 test scenarios in Auto Mode. (aireleasetracker.com / aiagentstore.ai)
• MAI-Cyber-1-Flash (microsoft) — New purpose-built cybersecurity model previewing Aug 3, packaged with the Project Perception agent system. (aiagentstore.ai)
• Dynatrace no-code Agent Builder — A tooling layer for spinning up custom observability/ops agents without code. (aiagentstore.ai)

🎓 AI Skills

• AI-skills demand spreads beyond tech — Bipartisan Policy Center's Skills Dashboard shows the fastest-growing industries for AI hiring are now employment/staffing agencies (+69% YoY), CPA offices (+55%), software publishers (+54%), and commercial banking (+51%). (Bipartisan Policy Center)
• What employers actually want — In accounting, microsoft-copilot (+85%), prompt engineering (+76%), and RAG for querying financial docs (+64%) lead; in banking, hugging-face proficiency (+77%), LLM skills (+68%), and generative AI (+67%) top the list. (Bipartisan Policy Center)
• Concentration risk remains — Three-quarters of all AI-skill demand still sits in just three occupational groups (computing/math, business/finance, management), a workforce-planning red flag. (Bipartisan Policy Center)

---

Sources: VKTR, Medium, Value Add VC, iwoszapar.com, Powerdrill, aiagentstore.ai, aireleasetracker.com, llm-stats.com, Bipartisan Policy Center. Compiled July 29, 2026.