In partnership with
Hey {{first_name | there}},
Since March, something has definitely changed. Claude has been hitting usage limits after 2 messages, sometimes even just one.
It’s annoying, and people are clearly frustrated (including me).
While I’m not sure what the exact reason is, one thing feels pretty obvious: this is intentional.
Anyways, there are still a few ways you can work around this. You won’t remove the limits completely, but you can stop hitting them so quickly.
Here’s what worked for me:
1. Never upload PDFs directly: They can consume up to 80% of your tokens in one go. Do this instead: give the PDF to ChatGPT, ask it to remove fillers and condense everything, then paste the cleaned version into Claude.
Use this prompt: “Read this document thoroughly. Remove all filler words, redundant text, and formatting. Extract only the core information. Return as condensed plain text with key points preserved.”
2. Use the Caveman method: Ask Claude to remove all filler and give short, direct answers.
Use this prompt:
“From now on, remove all filler words. No ‘the’, ‘is’, ‘am’, ‘are’. No pleasantries. No preamble. Direct answer only. Use short 3–6 word sentences. Run tools first, show result, then stop.”
3. Create a “COMPACT” skill: Long chats eat tokens fast. Ask Claude to summarize the entire conversation into 5–7 key bullet points, then start a new chat and paste that summary.
Use this prompt: “Create this skill: When I say ‘COMPACT’, summarize our entire conversation into 5–7 key bullet points including all critical context, decisions, and code. Format it for easy copy-paste into a new chat.”
4. Use Code Review Graph: This is a GitHub repo that turns your entire codebase into structured maps so Claude understands the architecture without reading every line. Saves a lot of tokens for coding related tasks.
5. Stop using Opus for everything: Opus burns way more tokens. Use Sonnet for most tasks (code, summaries, analysis) and Haiku for quick/simple tasks. Only use Opus for complex reasoning or critical work.
6. Use the session timing trick: Your usage window starts from your first message. Send a throwaway message a few hours before you actually start working so your reset happens during your workflow.
🧠 AI News of the Day
Perplexity just partnered with Plaid to let users connect their bank accounts, credit cards, loans, and investments directly into the platform. That means it can now analyze your spending, track your net worth, and even build custom dashboards or financial plans, all through simple prompts.
Why this matters: This is a big shift from “AI that answers questions” to “AI that manages your life.” Instead of using 3–4 different finance apps, you can now ask one AI things like “Am I overspending?” or “Can I afford this?” and get real, personalized answers. It’s basically moving toward an AI financial advisor for everyday users.
OpenAI is reportedly working on a new cybersecurity-focused AI product. But instead of releasing it publicly, it’s being rolled out to a small group of trusted partners only.
This comes right after Anthropic revealed its own highly capable (and potentially dangerous) model, Mythos, which is also being restricted due to its ability to find exploits and vulnerabilities.
Why this matters: We’re entering a phase where AI isn’t just useful, it’s powerful enough to be risky, and access itself is becoming a competitive advantage.
🛠️ AI Tools Worth Checking Out
Basquio: Turn spreadsheets into presentation-ready decks instantly.
AgentKit: Build AI-powered features and workflows for your product fast.
WebZum: Create and launch a complete website from a simple prompt.
ConceptSeek: Search inside videos and documents using meaning, not keywords.
New research suggests that tools like ChatGPT and Claude aren’t just helping you write, they might actually be shaping how you think, communicate, and even see the world.
The concern is something called “cultural homogenization”. Basically, AI tends to reflect a narrow set of perspectives (mostly Western, English-speaking), and over time, people using it start sounding and thinking more alike.
Why this matters: If billions of people rely on the same few AI systems, we could slowly lose diversity in ideas, culture, and problem-solving. AI isn’t just a tool anymore; it’s becoming a subtle influence on human thinking itself.
A detailed analysis by an AMD AI director claims that Claude Code’s performance has significantly degraded since early March. Across nearly 7,000 sessions, the data shows it now reads less context, rewrites more code unnecessarily, and abandons tasks mid-way at rates that were previously zero.
The suspected reason? Changes to how Claude handles “thinking”, specifically, reducing visible reasoning, which may have impacted how deeply it processes tasks. On top of that, rapid releases and multiple outages suggest the system might be struggling to keep up with growth.
Why this matters: If true, this explains why so many users feel like Claude has become unreliable. More importantly, it raises a bigger issue. AI tools can change overnight without clear communication. And when people depend on them for real work, that lack of transparency becomes a serious problem.
A newer model, GLM 5.1, is showing performance close to top-tier models like Opus. but at roughly one-third the cost in real-world tasks. It achieves this by aggressively using tools and optimizing how tasks are completed, making it one of the most cost-efficient options right now for agent-style workflows.
Why this matters: The AI race is no longer just about “who is smartest”, it’s about who is cheapest at scale. If models can deliver similar performance at a fraction of the cost, it could completely shift which tools developers and businesses choose going forward.
That’s it for today.
AI is moving fast, but the people who stay consistent with it are the ones who actually benefit.
See you in the next one.
- Aashish
What did you think about today's newsletter?
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning