@AIForPeople

Making AI practical for everyone. Agentic AI workflows, tools, experiments and open-source projects—explained with evidence, visuals Follow to learn AI

 Hyderabad, India
Joined July 2021
  • Tweets3.5K
  • Following56
  • Followers83
593Media
AI agents have learned to act. The harder skill is knowing when to WAIT. Microsoft Research’s SentinelBench tests agents on changing environments where progress can happen without agent action. 🧵
🤖 Made with AI
AIforPeople infographic explaining SentinelBench and why long-running agents should observe, wait, detect changes, then act.

ALT AIforPeople infographic explaining SentinelBench and why long-running agents should observe, wait, detect changes, then act.

1
11
more replies
Stretch the tasks to 40 minutes and the difference becomes dramatic. GPT-5.4 + wait_for: 69/100 tasks, $0.48 median cost. GPT-5.4 + sleep: 56/100 tasks, $4.65 median cost. That's 9.7× higher median cost for fewer successful tasks.
🤖 Made with AI
1
7
The deeper lesson: Autonomy is not continuous action. A strong long-running agent needs an ATTENTION POLICY: ACT. WAIT. DETECT. WAKE. STOP. The future agent may be defined as much by its waiting strategy as its reasoning model. Would you build agents around this loop?
🤖 Made with AI
2
Your software now has an agent user. Microsoft’s open-source Scope tests AI coding agents as users of APIs, SDKs, docs and tools—not just as code generators. That creates a new discipline: Agent Experience (AX). 🧵
AIforPeople infographic explaining Microsoft Scope and Agent Experience for AI coding agents.

ALT AIforPeople infographic explaining Microsoft Scope and Agent Experience for AI coding agents.

1
17
more replies
The fix surface is broader than prompts: → Documentation → SDK/API design → CLI behavior → MCP tools → Skills/instructions → Error messages Agent success is partly a product-design problem.
1
3
The deeper shift: Software is no longer designed only for humans to discover, understand and operate. AI agents are becoming users too. The winning products may be the ones that are easiest for BOTH humans and agents to use. Would you test your product this way?
2
AI security has a blind spot: the third-party software around the agent. Today’s Anthropic disclosure shows why. AI agents exploited weaknesses in connected business software and reached sensitive internal data. 🧵
AIforPeople infographic explaining why third-party business software is a security boundary for AI agents, using Anthropic's October 2026 disclosure as the case study.

ALT AIforPeople infographic explaining why third-party business software is a security boundary for AI agents, using Anthropic's October 2026 disclosure as the case study.

2
1
22
more replies
A stronger control stack: LEAST PRIVILEGE + short-lived credentials + third-party isolation + anomaly detection + rapid containment + patching + action logs Agents should get only the authority needed for the task.
1
2
The deeper lesson: AI-agent security is becoming a DEPENDENCY-GRAPH problem. The model may be well defended while a connected application becomes the bridge to sensitive data. Secure the entire graph—not just the model. Would your stack pass that test?
9
Load more