· Dennis Zimmer
Goldman restricts Claude Code as a 65B open model tops a benchmark
This week was about who gets to use coding agents and what those agents send out. Read it in this week's AgentMon Weekly.
New! AgentMon Start $15/device/month, 3 devices free for 6 months
Secure and compliant agentic development
Secure. Optimize. Automate AI in your enterprise.
Secure, automate and scale agentic productivity
- AgentMon Start Organization-wide visibility into AI agent activity, spend, security and device fleet
- Autonomous Security AI agents monitor, patch and automate your servers.
Hover or select a product to explore its activity
Live from AgentMon Start
4 events
- Cheaper model suggested agent-0F765 · 38% of its spend went to a premium model on tasks a smaller model completes; switching saves about $140 a week. OPTIMIZE
- Burn-rate spike alerted srv-0F769 · Agent spend 5× its 24h hourly average; team owner notified. ALERTED
- Prompt cache loss found host-0F76D · Cache hit rate down to 41% on 23 sessions; $63 lost this week because the system prompt keeps changing. REVIEW
- Task merged as-is agent-0F771 → payments-svc · Refund flow change passed CI and review with no human edits; $1.80 per success. MERGED
Illustrative, anonymized events
More Productive AI Agents. Better Output. Secure and Under Control.
32%
Output quality. Cite agent quality as a leading production barrier.
LangChain, 202645%
Productivity. Say oversight and rework consume over a quarter of expected productivity gains.
G2, 202652%
Security & control. Identify security, privacy or compliance as barriers to production.
Dynatrace, 2026
Used by teams at
Make AI Agents Productive, Reliable and Secure.
AI agents are becoming part of every organization. But do you know what they accomplish, what they cost, and what risks they introduce?
Keep tabs on the fast-spreading agentic networks in your organization
Where is data being leaked in your agentic networks?
Latest content
· Moshe
Innovation Should Not Require Permission
Anthropic's limits on game recreation in Claude Code could also block authorized remakes and legitimate work. A coding assistant should know the difference.
· Dennis
Scale without friction: review queues and long-running agents
Agents open changes faster than people review them and run for hours unattended. AgentMon Start shows review queues, file collisions and session timelines.
See what your AI agents are doing
AgentMon Start is free for 3 devices for 6 months. Paid plans start at $15 per device per month.