---
title: "From Prompts to Harnesses: Building Reliable AI Workflows"
description: "Amazon’s Anisha Malde joins Łukasz Chludziński to discuss harness engineering, agent skills, and how AI is changing cross-platform development."
url: https://www.callstack.com/podcasts/from-prompts-to-harnesses-building-reliable-ai-workflows
updated: 2026-11-10T12:30:03.172Z
---

# From Prompts to Harnesses: Building Reliable AI Workflows

Amazon’s Anisha Malde joins Łukasz Chludziński to discuss harness engineering, agent skills, and how AI is changing cross-platform development.

_Guest: Anisha Malde, Developer Advocate @ Amazon · Host: Łukasz Chludziński, Engineering Manager @ Callstack · Published 2026-11-10 · Category: AI_

Listen: [YouTube](https://youtu.be/1lohDYSgXIY) · [Spotify](https://open.spotify.com/episode/3RW28zA38g1wySkKtf7NX9?si=5WorBfhhQXSznlqur1Cg8w) · [Apple Podcasts](https://podcasts.apple.com/us/podcast/ai-agents-write-the-code-how-do-developers-stay-in/id1525543072?i=1000800989137) · [SoundCloud](https://soundcloud.com/user-985069846/harness-engineering-amazon)

Harness engineering involves designing the tools, context, and feedback around an AI model. These elements guide the agent’s work and give developers ways to investigate failures and intervene.

Recorded at Agent Conf 2026 in Warsaw, this episode of React Universe On Air brings together Anisha Malde from Amazon and our host Łukasz Chludziński. They trace the move from prompt engineering to context and harness engineering, connecting these ideas to familiar practices such as testing, task breakdown, and continuous integration.

Drawing on examples from [Amazon Devices Builder Tools (ADBT)](https://developer.amazon.com/docs/adbt/get-started), [Vega OS](https://www.developer.amazon.com/docs/vega/0.24/vega-get-started), and React Native, they discuss how platform knowledge becomes reusable agent skills. The conversation also examines how developers maintain their technical judgment as AI takes on more implementation work.

### What is harness engineering?

Anisha describes an agent as a model combined with a harness: the surrounding tools, context, and feedback that shape its behavior. She connects this to the test suites and engineering processes teams already use to guide development. Prompts remain part of that system.

Anisha identifies three goals for AI engineering: getting the model to do the intended job, doing it reliably, and being able to intervene and understand failures.

### AI-assisted development across TV platforms

Amazon Devices Builder Tools combines tools exposed through the Model Context Protocol (MCP) with agent skills and workflows for Amazon device development. Anisha discusses building React Native apps for Vega OS and migrating existing apps from Fire OS, alongside community efforts to share knowledge across platforms. She also explains where platform-specific requirements still need attention.

### How does platform knowledge become an agent skill?

Knowing which skills and tools to give an agent requires knowledge of the platform and the task. Anisha and Łukasz discuss how individual expertise can become shared guidance, from TV focus management to React Native best practices.

Anisha describes a [community agent skill for moving from Vega to Expo](https://github.com/AmazonAppDev/devices-agent-skills) and explains how she combines Amazon’s tooling with [Callstack’s React Native best-practices skills](https://github.com/callstackincubator/agent-skills/tree/main/skills/react-native-best-practices). They also consider which workflows teams can standardize and where developers need room to adapt.

### Keeping technical judgment as AI takes on more coding

Łukasz raises a concern about maintaining platform expertise when developers spend less time working directly with APIs and implementation details. Together, they explore the tension between abstracting away routine work and retaining enough knowledge to diagnose unfamiliar bugs and choose the right tools.

### Applying engineering practices to agent workflows

Task breakdown, specifications, and CI/CD return throughout the discussion. Anisha draws parallels between these established practices and the context, token budgets, and harnesses used with agents. They close by considering how AI can give developers more space to focus on the experience they want users to have.

Watch the episode to learn how harness engineering and reusable agent skills fit into development workflows, and what developers need to stay in control.

## Resources

- Anisha Malde on [LinkedIn](https://uk.linkedin.com/in/anishamalde), [GitHub](https://github.com/anishamalde), and [X](https://x.com/anisha_malde)
- Łukasz Chludziński on [LinkedIn](https://www.linkedin.com/in/lukasz-app) and [X](https://x.com/lukasz_app)
- [Amazon Devices Builder Tools for AI: getting started](https://developer.amazon.com/docs/adbt/get-started)
- [Vega OS developer documentation](https://www.developer.amazon.com/docs/vega/0.24/vega-get-started)
- [Amazon Devices community agent skills](https://github.com/AmazonAppDev/devices-agent-skills)
- [Callstack’s React Native best practices agent skill](https://github.com/callstackincubator/agent-skills/tree/main/skills/react-native-best-practices)
- [OpenIAP: getting started](https://www.openiap.dev/docs/getting-started)
- [Model Context Protocol documentation](https://modelcontextprotocol.io/docs/getting-started/intro) (further reading)
- [Agent Skills overview](https://agentskills.io/home) (further reading)

Insights

## Learn more about AI

Here's everything we published recently on this topic.

[View all](https://www.callstack.com/insights)

- [Efficient On-Device LLMs in React Native Done Right](https://www.callstack.com/events/efficient-on-device-llms-in-react-native-done-right) (Nov 17 · Talk): Artur Morys-Magiera will cover efficient on-device LLMs in React Native, including runtimes, model selection, quantization, and platform constraints.
- [On-Device AI After WWDC: What Changed for React Native](https://www.callstack.com/events/on-device-ai-after-wwdc-what-changed-for-react-native) (Nov 17 · Talk): Mike Grabowski will explain what changed for system-provided on-device AI after WWDC and what React Native developers can build with it.
- [Beyond Coding Agents: Making ADLC Work Across Platforms](https://www.callstack.com/events/beyond-coding-agents-making-adlc-work-across-platforms) (Nov 16 · Talk): Mike Grabowski will explain how to close the full agentic development lifecycle across mobile, web, and native platforms.
- [Building the React Native Software Factory](https://www.callstack.com/events/building-the-react-native-software-factory) (Nov 11 · Talk): Join us at React Foundation Meetup London on November 11 to explore how AI agents build and debug React Native apps, from one terminal to the cloud.
- [Build, Launch, and Maintain: The Business Case for a Software Factory](https://www.callstack.com/blog/build-launch-and-maintain-the-business-case-for-a-software-factory) (Nov 10 · Article): Explore an anonymized software factory business case, its delivery workflow, and the staffing assumptions behind a projected 65% reduction in total costs.
- [Building a Software Factory: Write Specs an Agent Can Build From](https://www.callstack.com/blog/building-a-software-factory-write-specs-an-agent-can-build-from) (Nov 6 · Article): Learn how to research requirements, resolve open decisions, and write testable specs that AI coding agents can implement without the planning conversation.
- [ Large-Scale Migrations With Coding Agents: Planning, Verification, and Rollback](https://www.callstack.com/blog/large-scale-migrations-with-coding-agents-planning-verification-and-rollback) (Nov 4 · Article): Plan codebase migrations with coding agents using reviewable requirements, verification checks, staged rollouts, and defined rollback criteria.
- [Quantization Impact on LLM Quality: A Case Study of Apex](https://www.callstack.com/events/quantization-impact-on-llm-quality-a-case-study-of-apex-a-gemma-based-model) (Nov 4 · Talk): Artur Morys-Magiera and Lech Kalinowski share findings from 344 benchmark runs with Apex, our Gemma-based LLM, examining the trade-offs between model size, decode speed, and quality.
- [Conversational Banking: From Customer Questions to Authorized Actions](https://www.callstack.com/blog/conversational-banking-from-customer-questions-to-authorized-actions) (Nov 2 · Article): Explore conversational banking architecture through a card-dispute example, covering tool access, authorization, execution, observability, and human handoff.