Building AI Coworkers: How Neople Is Making Agents Work Where You Work

This podcast episode features three leaders from Neople, a company building 'digital coworkers' that automate work such as customer support, invoice processing, and drafting emails. The conversation traces how Neople moved from simple AI response suggestions to fully autonomous agents, and how its conversational workflow builder lets non-technical users create automations. A central theme is quality: the team describes evals that run in production to catch hallucinations before messages reach customers, and feedback loops that turn customer input into eval pipelines. The episode is useful for product teams wondering how to balance deterministic code, agents, and guardrails when shipping AI features.

01Key takeaways

• Balance deterministic code, agents, and guardrails rather than assuming an LLM alone can handle every task.
• Run evals in production to catch hallucinations before outputs reach end customers.
• Treat customer feedback as a source of eval cases that continuously improve product quality.
• Teach non-technical users to decompose tasks while they build automations conversationally.
• Expand AI teammates gradually, moving from suggestions to autonomous actions as trust is earned.

02Key sections

Origin and product vision

The guests introduce Neople's concept of digital coworkers that combine the reliability of automation with the flexibility of AI. The company's origin story frames the product as a teammate rather than a tool.

Evolution from suggestions to automations

Neople's product progressed from suggesting responses to running fully autonomous customer service agents. The team describes the tradeoffs of letting AI act versus keeping humans in the loop.

Balancing code, agents and guardrails

The team explains moving away from the belief that LLMs will solve everything, toward a mix of deterministic code, agentic behavior, and guardrails that constrain outputs.

Evals and quality in production

Neople runs evaluations in production to detect hallucinations before emails reach customers, and incorporates customer feedback into eval pipelines that improve quality over time.

Conversational workflow building for non-technical users

Non-technical users build automations through conversation, and the product teaches them task decomposition along the way. Knowledge retrieval, embeddings, and integrations support this workflow.

03From the post

“Listen to this episode on: Spotify | Apple Podcasts What if your next teammate was an AI coworker — one that could answer support tickets, process invoices, or even draft your next email — and your non-technical colleagues could teach it how to do those tasks themselves? In this episode, host Teresa”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.