I Fixed Claude's Token Limits. Here's How.
The speaker hit Claude's usage limits despite having a $200/month plan and shows how to optimize AI agent setups by using different models for different agents. They demonstrate switching the main orchestrator to Sonnet while keeping specialized agents on Opus, and adjusting effort levels to reduce token consumption.
Summary
The speaker begins by explaining their frustration with constantly hitting Claude's usage limits, even with a premium $200/month plan that includes weekly limits. They note this is a recent change, as they previously ran everything on the Opus model without any issues. The speaker shows their usage dashboard, revealing they're close to session limits and have already spent an additional €70 on top of their plan. They attribute this to being 'very generous' with their AI agents and model usage. The main solution involves optimizing their multi-agent setup within Claude. The speaker demonstrates how to access and modify individual agents through the /agents command, showing that each agent can be configured to use different models rather than inheriting from the parent session. Their strategy involves using Sonnet for the main orchestrator agent 'Larry,' whose job is simply to understand which team member is best for a task and delegate accordingly. For specialized work like coding, they keep specific agents like 'Felix' on the more powerful Opus model. Additionally, they show the new effort functionality that can be adjusted with arrow keys, explaining they were previously running Opus with high effort and a 1 million context window, which contributed to excessive token usage. By switching to Sonnet with medium effort for coordination and reserving Opus for specialized tasks, they can significantly reduce costs while maintaining functionality.
About this episode
Stop burning through your Claude Max plan in 3 days. 🚀✨ Build your Future-Proof AI Productivity System TODAY: https://myicor.com If you're running a multi-agent AI team on Claude Code, you've probably hit the same wall I did. This video shows the exact fix. I was running all 20+ agents on Opus 4.6 with high effort and a 1 million token context window. Session limits, weekly limits, extra charges on top of the $200/month Max plan. The solution: assign the right model tier per agent. My orchestrator now runs on Sonnet at medium effort. Specialists like Felix still get Opus for the heavy lifting. I also walk through the /agents UI and the new effort control (arrow keys) so you can do this yourself in 5 minutes. TIMESTAMPS 0:00 Hitting the wall: session and weekly limits 0:27 What changed: more tokens, more obvious problem 0:52 My actual plan usage (70 euros overage) 1:20 The fix: optimizing model assignments 1:38 Inside the Larry folder and Claude Code agents 1:58 The /agents UI: viewing and editing agent models 2:45 Switching Quinn to Haiku for lightweight tasks 3:29 Switching Larry to Sonnet 4.6 with effort control 4:22 Why I was burning tokens: Opus + high effort + 1M context 4:39 The new setup: Sonnet orchestrator, Opus specialists 5:17 How to apply this to your own AI team RESOURCES Claude Code: https://claude.ai/code Anthropic Max Plans: https://www.anthropic.com/pricing SUBSCRIBE for more AI productivity walkthroughs every week. ABOUT Tom helps professionals build AI productivity systems that work, using the tool-agnostic ICOR Methodology, courses, personal coaching, and a growing community on myICOR. LinkedIn: https://www.linkedin.com/in/tomsolid/ X: https://x.com/TomSolidPM Podcast: https://www.youtube.com/@productivitylikeapro Disclaimer: Some links may be affiliate links. If you click and purchase, I may earn a small commission at no extra cost to you. #ClaudeAI #claudecode #AIProductivity #claudecowork #myICOR #ICOR #ProductivitySystem #BusyProfessionals
Key Insights
- The speaker was previously running Opus 4.6 with a 1 million context window in high effort mode constantly, which explains why they started running out of tokens
- Claude automatically creates agents at the project level when you set up AI teams using simple folder structures with instructions
- The main orchestrator agent Larry only needs to understand who is the best team member for a job and delegate to them, making Sonnet sufficient for this coordination role
Topics
Transcript
[0:00] Like many of you, I also started to constantly hit my limits and now even running out of my weekly limits with my max model 20x plan with $200 per month. Well, yes, I'm do I'm using my AI team for many things for my PKA, personal knowledge assistance, but also running the business team here doing a lot of things like coding and so on. So, I'm really leveraging this nonstop. Nonetheless, there's a total shift because I never hit any limits in the [0:32] past and I ran everything on OPUS model all the time and no issues whatsoever. I never hit these limits. So, there's certainly something that changed in the background. So, here is actually…
Full transcript available for MurmurCast members
Sign Up to AccessMore from ICOR with Tom | AI Productivity
Claude sucks. And here is why.
This video explores why Claude and other AI models produce inconsistent results, examining how model selection (Haiku vs Sonnet vs Opus vs Claude 3.5 Fable), effort levels, prompt clarity, and system context dramatically affect output quality. The speaker demonstrates these differences through a Venn diagram creation task and advocates for using organized folder systems with documented SOPs and code-based solutions to ensure consistent AI performance.
Claude can now prompt its own SESSIONS. I built the boss.
Claude now has the ability to interact with multiple sessions, allowing a streamlined orchestration of tasks through an agent called Larry. This integration can simplify task management and enhance productivity by enabling a single main session to oversee various sub-sessions or tasks.
I switched off Claude's memory. My folder does it better.
Tom explains why he disabled Claude's native memory feature and instead built a local folder-based system (myPKM) to maintain persistent context with AI agents. He shares an audit revealing systemic issues like contradictory documentation, forgotten rules, and parallel session conflicts, then outlines how to systematically fix these problems using the ICOR productivity methodology.
I do not download invoices anymore. My folder and Claude do it.
The speaker demonstrates how he uses AI (Claude) integrated with his local folder system to automate invoice collection for his accountant, downloading 36 invoices in 5 minutes with minimal manual intervention. He explains his AI team methodology, work streams, and how persistent local knowledge bases enable AI to handle recurring tasks independently without constant re-explanation.
Claude finished the work. Then I lost it in the chat.
A creator demonstrates a comprehensive local-folder-based system for managing AI-generated deliverables asynchronously, featuring an interactive browser interface that tracks decisions, changes, and conversations without relying on external databases or web applications. The system enables collaborative editing, visual annotations, and automatic context preservation across multiple parallel AI projects.