Skip to main content
bash TV

Agents That Write Their Own Tools at Runtime — Sandhya Subramani, AWS

AI Engineer

8.7K views4 Oct 2026

YouTube

This agent starts with zero tools. Ask it for something it can't do, and it writes the tool and uses it without restarting. Sandhya Subramani, senior developer advocate for GenAI at AWS, demos meta-tooling with the open-source Strands Agents SDK. That means agents that write, load and use their own tools at runtime, and even spin up their own sub-agents. All it takes is a system prompt describing what a good tool looks like, plus three tools: editor, shell and load_tool. She walks through agentic patterns (swarm, graph, handoff, workflow), demos a travel planner that builds its own sub-agents, and covers the evals and guardrails you need before trusting a self-modifying agent in production. In this talk: • Meta-tooling: editor, shell and load_tool plus one system prompt • Agents that create their own sub-agents • Evals for self-modifying agents: goal success, tool choice, parameters, inter-agent flow • Guardrails: sandboxed execution, constrained permissions and observability SPEAKER Sandhya Subramani, Senior Developer Advocate, GenAI, AWS LinkedIn: https://www.linkedin.com/in/sandhyasubramani/ GitHub: https://github.com/sandhya-subramani LINKS Strands Agents: https://strandsagents.com Strands Agents (GitHub): https://github.com/strands-agents Strands Agents tools: https://github.com/strands-agents/tools CHAPTERS 0:00 Agents that forge their own tools 0:40 Fixing code at runtime 1:24 Demo: an agent with zero tools 2:49 It writes a calculator tool 3:39 It writes a character counter 4:59 Meta-tooling with Strands Agents 5:59 What Strands Agents is 6:44 Three tools: editor, shell, load_tool 7:43 Teaching the agent what a good tool looks like 8:43 The system prompt in code 9:53 Why this matters in production 10:43 Agents that create agents 11:03 Swarm, graph, handoff, workflow 11:53 Demo: a self-assembling travel planner 14:17 Self-healing agents 15:21 With great power: evals 17:06 Guardrails: sandbox, permissions, observability 18:56 Self-improving agents 19:21 Strands wrote its own TypeScript version Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #AIAgents #AWS #AIEngineer

Join the discussion

Sign in to join the discussion

Sign in