Also, this webinar was kind of fun, because WE also asked YOU some questions! 
Like…
What AI tool(s) are you using to supercharge your work as a developer?
What’s the most annoying thing about coding AI agents right now?
- Troubleshooting: How interruptions are handled because of infra issues; Going in loop without making progress. Realizing when it has hit a roadblock and seeks human input; They get stuck in ridiculous annoying loops when biting off too much to chew on; Hitting a loop in conversations and getting no where in some cases; Hitting context limit
- Agent Reliability Hallucinations/stochasticity/non-determinism; variability in LLM responses for the same inputs; Knowing if I made the system better or worse with each iteration of prompt/code; They’ll never push back, never say “you ask a stupid thing, I won’t do that”; testing
- Prompt Engineering: Preparing training data; Creating prompts, Writing prompts to validate AI results
- Goal Setting: Using a conversation to address intermediary goals that may be conditionally required; Addressing an additional goal(s) before accomplishing our primary goal.
- Observability and Evaluation: Visualization of the flow for past runs; How to ensure AI responses are correct;
- Integration: Integrating the AI agent into our UI/UX built-in product; Managing different tools per LLM from different brands
- Improved Temporal AI Features Would like to see more support from Temporal in the Python SDK for supporting agents and chat.
- Pace of Innovation * Having to learn a lot, and kind of continue doing what i am doing and actively up-skilling to pick up the new techs; Rapidly changing stuff before i can think of building something comes up; A lot of folks don’t think about the paradigm of execute ↔ review ↔ commit enough and it makes it harder to get buy-in and move fast
- Little Guidance Lack of documentation, credible knowledge base to rely on; Not having enough resources or tools to know how and where to get things started with AI development.
- Orchestration everything is easy, except for the orchestration; Workflow orchestration 100%
- Deployment Deployment to EKS
- Proliferation of Choice so many frameworks, sdks, no consolidation
- Massive Implications Because of this you can do so much but because you can do so much, there is also so much that can go wrong/be exploited; So much power (good tone), but also …so much power
What use cases are you excited about with AI?
What Temporal + AI topic would you like us to talk about next?
- RAG
- MCP
- AI Orchestration
(Credit to Word Cloud Generator for the Word Clouds!)