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
- Tweets3.6K
- Following56
- Followers83
- Likes44
Giving an AI agent tools doesn’t automatically make it useful. It creates a new failure surface. ⚙️
A tool call can be syntactically valid and still be wrong, unsafe, or based on bad assumptions.
Here’s how to design reliable tool use. 🧵
ALT AIforPeople infographic on reliable AI tool calling: clear schemas, input validation, permission checks, and result verification. A valid call is not the same as a successful outcome.
Official guide: OpenAI’s function-calling best practices cover clear names, parameter descriptions, predictable schemas, and strict mode.
developers.openai.com/api/do…
Also useful: Structured Outputs can constrain outputs to a schema, but application logic still needs to handle errors and validate real-world outcomes.
developers.openai.com/api/do…
Your AI agent doesn’t need a bigger context window as much as it needs better context. 🧠
Context engineering is the art of giving a model the right information at the right time—not dumping everything into the prompt.
Here’s the practical playbook. 🧵
ALT AIforPeople infographic: Better context beats bigger prompts. Four steps for AI agents: retrieve what matters, rank by relevance, compress stale history, and verify against sources. More tokens do not equal more signal.
Further reading: Microsoft's practical context-engineering guide
code.visualstudio.com/docs/a…
Further reading: OpenAI's guide to compaction for long-running agent workflows
developers.openai.com/api/do…
