> Markdown version of [/videos/100547-building-ai-products-vs-building-with-ai](https://www.wearedevelopers.com/videos/100547-building-ai-products-vs-building-with-ai). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Building AI Products vs. Building With AI

With code generation costs approaching zero, manual pull request reviews are becoming obsolete. Learn how engineers are evolving into operators of high-velocity AI software factories.

- **Speakers:** [Aparna Dhinakaran](https://www.wearedevelopers.com/@aparna-dhinakaran), [Rukmini Reddy](https://www.wearedevelopers.com/@rukmini-reddy), [Tamar Bercovici](https://www.wearedevelopers.com/@tamar-bercovici)
- **Event:** World Congress 2026 North America
- **Published:** September 25, 2026
- **Duration:** 30:15
- **URL:** https://www.wearedevelopers.com/videos/100547-building-ai-products-vs-building-with-ai

## Summary

As AI adoption transitions from early experimentation to ubiquitous reality, engineering teams face a new challenge: managing the unprecedented volume of AI-generated code running in production. The distinction between building AI products for customers and using AI internally to build software is blurring. With the cost of code generation approaching zero, the half-life of software has drastically decreased, forcing teams to constantly rebuild and adapt. This paradigm shift elevates the engineer’s role from writing boilerplate to operating an internal software factory, focusing intensely on user experience, system architecture, and product reliability.

To support this accelerated development cycle, organizations must invest heavily in dynamic agent harnesses and infrastructure guardrails. Managing token spend has replaced traditional headcount budgeting, requiring shared platforms that track usage and provide proper context so agents avoid starting from a blank slate. Because AI introduces non-deterministic outcomes, traditional unit testing is no longer sufficient; engineers must adopt evaluation frameworks to measure agent quality. Rather than abandoning failed LLM outputs, successful teams actively troubleshoot by injecting the correct context, architecture documents, and tools to improve agent performance.

The sheer velocity of AI code generation means human reviewers can no longer manually audit every pull request or monitor all telemetry data. Consequently, organizations are deploying AI agents to conduct active observability, reviewing logs and traces to catch unknown anomalies that human operators would inevitably miss. As agentic workflows scale, the classic build versus buy debate has evolved. While the cost of building custom internal tools has dropped, relying on established vendors for specialized domains like compliance, security, and five-nines reliability remains crucial. Ultimately, teams must avoid the trap of throwing the largest, most expensive model at every problem, instead right-sizing their infrastructure to maintain enterprise-grade trust.

**Keywords:** ai-generated code management, software development lifecycle, agentic workflows, production observability, token spend management, LLM evaluation frameworks, human-in-the-loop automation, non-deterministic testing, engineering platform guardrails, build versus buy strategy, enterprise AI adoption, ai product development, automated telemetry review, dynamic agent harnesses, infrastructure budgeting

## Chapters

1. **Transitioning from AI adoption to team orchestration** (02:12) — To move beyond basic model access, engineering teams must orchestrate AI tools to achieve massive productivity gains.
1. **Rebuilding products and creating the AI software factory** (04:52) — Because the cost of code generation has plummeted, teams can rapidly rebuild architectures and use agents to automate telemetry analysis.
1. **Implementing guardrails and tracking AI token spend** (07:46) — To manage the risks of decentralized AI usage, teams need a shared platform that enforces guardrails and tracks token spend.
1. **Evaluating the quality of non-deterministic AI product features** (10:17) — Since AI assistants generate highly personalized outcomes, product engineers must adopt new evaluation frameworks for non-deterministic features.
1. **Developing adaptive harnesses for dynamic model routing** (12:59) — To balance price and performance, engineering organizations must develop adaptive harnesses that dynamically route workloads to the optimal model.
1. **Deploying agents for automated observability and incident response** (15:21) — Because humans cannot keep pace with AI-generated code, organizations should deploy agents to review logs and uncover unknown system vulnerabilities.
1. **Scaling production triage to manage AI-generated code volume** (19:39) — As development volume surges, engineers must implement systemic reinforcement loops to troubleshoot failures and improve agent outputs.
1. **Navigating build versus buy decisions for AI infrastructure** (23:44) — To ensure customer reliability, organizations should only build custom AI infrastructure when it directly supports their core business path.

## Related Moments

- [Balancing AI investment with product shipping commitments](https://www.wearedevelopers.com/videos/100559-beyond-the-code-human-ai-synergies-in-product-development) (from "Beyond the Code: Human-AI Synergies in Product Development")
- [Transitioning to AI-native engineering teams and workflows](https://www.wearedevelopers.com/videos/100508-legacy-as-a-launchpad-how-yahoo-mail-is-undergoing-a-product-and-engineering-transformation) (from "Legacy as a Launchpad: How Yahoo Mail is Undergoing a Product and Engineering Transformation")
- [Shifting developer workloads and realistic AI productivity gains](https://www.wearedevelopers.com/videos/1830-wearedevelopers-live-speculaitions) (from "WeAreDevelopers LIVE - SpeculAItions")
- [Shifting from implementation to deciding what to build](https://www.wearedevelopers.com/videos/100543-signal-layer-what-to-build-when-anything-can-be-built) (from "Signal Layer: What to Build When Anything Can Be Built")
- [Transitioning software engineering teams to AI-native development workflows](https://www.wearedevelopers.com/videos/100087-ai-ready-what-enterprise-transformation-actually-takes) (from "AI-Ready? What Enterprise Transformation Actually Takes")
- [Scaling engineering teams and identifying evolving delivery bottlenecks](https://www.wearedevelopers.com/videos/100443-engineering-moneyball-how-we-benchmarked-google-vs-meta) (from "Engineering Moneyball: How We Benchmarked Google vs Meta")

## Related Articles

- [Exploring AI: Opportunities and Risks for Developers](https://www.wearedevelopers.com/magazine/522-exploring-ai-opportunities-and-risks-for-developers)
- [What is Software Engineering in the Age of AI?](https://www.wearedevelopers.com/magazine/640-what-is-software-engineering-in-the-age-of-ai)
- [Transforming Software Development: The Role of AI and Developer Tools](https://www.wearedevelopers.com/magazine/527-transforming-software-development-the-role-of-ai-and-developer-tools)
- [What is Agentic Programming and Why Should Developers Care?](https://www.wearedevelopers.com/magazine/625-what-is-agentic-programming-and-why-should-developers-care)

## Related Jobs

- [Senior Product Manager](https://www.wearedevelopers.com/jobs/48353-senior-product-manager) at **PagerDuty**
- [Senior AI/ML Engineer](https://www.wearedevelopers.com/jobs/48352-senior-ai-ml-engineer) at **PagerDuty**
- [Senior AI Developer](https://www.wearedevelopers.com/jobs/ext/2836034-senior-ai-developer) at **PwC**
- [Partner Sales Director - AI Alliances - Model Providers](https://www.wearedevelopers.com/jobs/48429-partner-sales-director-ai-alliances-model-providers) at **Dynatrace**
- [Principal Software Engineer (m/f/x)](https://www.wearedevelopers.com/jobs/48548-principal-software-engineer-m-f-x) at **Dynatrace**
- [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat**