Agentification & platforms
Key Questions
What agent platforms are expanding?
Platforms like OpenClaw, MCP, Google Agent Executor, Kimi Work, and Vercel HarnessAgent are growing rapidly. TrueFoundry acquired Seldon AI amid 78% pilot and 14% production rates.
Why is persistent memory a key weakness for agents?
Persistent memory remains a major limitation in agent systems. Solutions like YourMemory 2.0 use MCP-native approaches and Ebbinghaus forgetting curves to address it.
What governance challenges do agentic AI face?
McKinsey and Deloitte stats show 88% adoption but only 25% in production. Articles emphasize human-in-the-loop needs after reversals at Ford and CBA.
How are agent gateways evolving?
Agent gateways are consolidating as the control plane, with Palo Alto acquiring Portkey and Solo.io donating to the Linux Foundation. They address security and orchestration.
What is Platform Engineering 2.0?
Platform Engineering 2.0 focuses on AI-native, multi-persona, FinOps, security, and composable pillars. It responds to governance and cost challenges from agent demands.
What observability tools support agents?
Heron provides passive eBPF observability. ThreatModeler Nexus offers agentic threat modeling while Octolens supplies an agent data layer via MCP.
What enterprise initiatives support agent deployment?
Microsoft launched a $2.5B Frontier Company with 6,000 experts. Operationalizing Agentic AI frameworks map governance maturity to autonomy stages.
How do startups target agent opportunities?
Greg Eisenberg notes opportunities in spend controls, sandboxes, and agent-to-agent negotiation. Stanford's agent-native Git addresses state management for long runs.
Agent platforms expanding: OpenClaw, MCP, Google Agent Executor, Kimi Work, Vercel HarnessAgent. TrueFoundry acquires Seldon AI — 78% pilots, 14% production. Persistent memory remains key weakness. Deloitte embeds multi-agent AI into audit workflows. Lenovo Hybrid AI Advantage claims 8-18x lower cost per token on-premises. Heron offers passive eBPF observability. McKinsey/Deloitte stats: 88% use, only 25% production. SquidHub for collaborative human-agent rooms. ThreatModeler Nexus for agentic threat modeling. Moving AI Agents From Pilot to Production article provides concrete playbook. Cyber AI tool safety debate. YourMemory 2.0 (MCP-native, consolidation via Ebbinghaus forgetting curve) addresses persistent memory gap. Data strategy article reinforces data-first approach. Enterprise AI platform article reframes bottleneck from GPUs to software process layer. AI Tokenomics article highlights token-based pricing reshaping enterprise AI strategy, agentic AI token explosion, and lack of mature ROI frameworks — reinforcing cost governance concerns. Aurora Mobile unveils omni-channel AI solutions at HKPC with government subsidies. Vercel Sandbox adds Docker and FUSE support for unconstrained agent runtimes — major infrastructure improvement. Claude Science autonomously analyzed sequencing data, wrote a paper, submitted, got rejected, and applied for jobs in 8 hours — demonstrating agentic scientific workflow. Enterprise AI Security article highlights agent blind spots (data exfiltration, permission bypass). Article 'Enterprise AI Deployments Stumble When Companies Rush Automation Without Human Expertise' — Ford and CBA reversals confirm human-in-the-loop necessity. Enterprise AI's center of gravity shifts from models to orchestration, governance, and ROI — practitioner quotes reinforce governance as velocity layer, not brake. CFO scrutiny and measurement tied to decision quality timely for Q3 budgets. Agent gateways consolidating as control plane for enterprise AI — Palo Alto buying Portkey, Solo.io donating to Linux Foundation, reinforcing governance shift. Stanford researchers build agent-native Git for state management during long agent runs — addresses context tax and remediation debt. Octolens offers agent data layer via MCP/webhooks for external awareness. Operationalizing Agentic AI framework maps governance maturity to autonomy stages. Microsoft $2.5B Frontier Company with 6,000 experts signals implementation over model quality. Devin Desktop as agent command center. FLARE-AI fills vulnerability disclosure gap. Platform Engineering 2.0 pillars (AI-native, multi-persona, FinOps, security, composable) directly address governance and cost challenges. Greg Eisenberg thread on building startups for agents (spend controls, sandboxes, agent-to-agent negotiation) reinforces opportunity.