# Zach Lloyd

Zach Lloyd writes about building software factories: systems that use AI agents and supporting infrastructure to increase software development throughput.

- Profile: https://b2binfluencer.marketing/creators/zach-lloyd/
- LinkedIn: https://www.linkedin.com/in/zachlloyd
- LinkedIn followers: 18,744
- Topics: artificial-intelligence
- Based on 12 dated posts, latest 2026-10-16

## Summary

Zach Lloyd is most useful for explaining software factories through layered infrastructure, practical metrics, and tool integration examples.

- Best for: Developer-tools brands running technical webinars or architecture guides on software factories.
- Not for: Consumer wellness campaigns needing short-form video, lifestyle content, or guaranteed reach.
- Check first: Confirm audience composition, reach and engagement data, media kit, rate card, and sponsorship formats.

### Takeaways

- A software factory should be an open, composable stack defined in code.
- Judge a factory by throughput, measured in human touches per pull request.
- Integrations are configuration, and work should start where people already talk.

## Main ideas

- **A software factory should be an open, composable stack defined in code**: Lloyd lays out software factories as layered infrastructure, with a job for each layer: factories-as-code at the base; a context layer (internal MCPs and CLIs, agent memories, conversation and access logs, agent rules and skills); a compute layer of remote, pausable environments; an inference layer that swaps between frontier and open-weight models and different agent harnesses; improvement infrastructure; orchestration for launching agents and steering live sessions; and access points through UIs, a direct MCP, and third-party surfaces such as code forges. They define three terms directly: open means the stack works with any model, agent, and hosting configuration; composable means a team can adopt part or all of it and use the pieces however it wants; defined in code means factory state is versioned, testable, and revertible. The aim is to let a team tune quality, cost, and speed over time without rebuilding the whole system. Versioning makes benchmarking, A/B testing, and rollbacks possible.
- **Keep the data and ability to verify inside your own walls**: Two conditions run through Lloyd's stack description. The first is data control: if a team uses a third-party context layer, it should be able to store all that layer's data on its own infrastructure, no outside provider should train on it, and zero data retention is a must. The context layer holds agent memories, skill files, conversation and access logs, and rules—the raw material the factory runs on and the record of how it behaves. The second is verification: Lloyd wants computer-use models connected to remote development environments to test the apps agents build, enabling reproduction, verification, and prototyping. Together, these conditions make a team's ability to inspect and trust what its factory produces central to its usability.
- **Judge a factory by throughput, measured in human touches per pull request**: Lloyd frames the shift in software engineering as a move from building the product to improving the factory, with a single efficiency measure to watch: human touches per pull request, which should fall over time. They also name the wider set: DORA-style metrics such as cost per PR, and qualitative measures like code quality and efficiency scored by an LLM acting as a judge. The numbers close the improvement loop. Agents find their own failure points and suggest skill improvements, while test suites with varied configurations measure how model and harness combinations perform on real work. Measurement becomes a recurring practice, not a one-off report.
- **Integrations are configuration, and work should start where people already talk**: Lloyd's Microsoft Teams and Azure DevOps announcement applies the code-defined pattern to enterprise surfaces: integrations, agents, triggers, and automations are declared as version-controlled configuration rather than clicked into a console. The flexible-infrastructure argument is that agents should be able to run from anywhere. They detail the mechanics: mentioning agents from work items and pull requests, triggering automations from work item and PR events, giving each factory its own identity for Git and pull request operations, working across selected repositories with scoped access, and shipping pull requests back through Azure DevOps. In Teams, users start by mentioning the app in an existing channel or thread; it keeps context as work moves from conversation to implementation and follows up in-thread while agents work. The post is also a vendor announcement: it positions the product as first in the industry to support the full Microsoft suite, offers early access and up to $10,000 in free factory usage for eligible teams, and bases its strongest claim on the company's own statement. Teams evaluating the open-stack argument should ask whether the Microsoft suite remains an optional piece of a stack they control.
- **Adoption is phased and cultural, not just technical**: Lloyd advertised a panel with Adam Huda, a senior engineering manager at Uber, and Christopher Handel, director of product for the AI platform at Faire, to compare notes on adopting software factories. The announced agenda is operational rather than conceptual: what each phase looks like in practice and how to tell whether you are in crawl, walk, or run; how to architect the factory stack across runners, environments, scorers, benchmarks, and quality gates; when to build versus partner; which metrics to track, including human touchpoints per change, cost per unit of software shipped, and governance signals; and the cultural changes needed to make the system stick.

## Work to read

- [Software factories as an infrastructure stack](https://www.linkedin.com/posts/zachlloyd_software-factories-should-be-built-on-an-activity-7508555853419982848-mBko)
- [Warp Factories adds Microsoft Teams and Azure DevOps](https://www.linkedin.com/posts/zachlloyd_warp-factories-now-supports-microsoft-teams-activity-7513661841932308481-hecM)
- [Factory engineering and human touches per PR](https://www.linkedin.com/posts/zachlloyd_software-engineering-is-shifting-to-factory-activity-7511454989521977345-gDqO)
- [Event: Comparing notes on adopting software factories](https://www.linkedin.com/posts/zachlloyd_moving-from-interactive-agents-to-software-activity-7511390825260670977-qHdj)

## Brand relationships

- **Warp** (employment)
- **Microsoft** (disclosed_partnership)
- **Uber** (mention)
- **Faire** (mention)

## Partnerships

Zach Lloyd’s posts in this evidence set announce products and events for Warp’s software-factory offering; they do not present creator sponsorship packages. Across the four posts, there is no media kit, rate card, sponsored format, disclosure practice, or official partnership inquiry route. The commercial signals are Warp’s Microsoft Teams and Azure DevOps integrations, an early-access offer of up to $10,000 in free factory usage for eligible teams, and a panel session with an engineering manager from Uber and a director of product from Faire. For a technical webinar or architecture guide collaboration, treat these as evidence of what they promote and who they’ll share a stage with—not as standing partnership terms. Confirm scope, deliverables, and pricing directly before making plans.

## Alternatives

- Cole Medin: Choose Cole Medin for the same developer-tools thought-leadership brief when the campaign should show what it takes to run an AI software factory, not just describe its layers. The available posts describe a working loop: a GitHub issue in, a pull request out, each lap building one issue, testing the app as a user, reviewing code, and merging to staging only when CI is green. They also cover release planning with 10–20 tickets, holdout test scenarios the builder never sees, a manager agent that pings when work is stuck, and human batch review in staging. That makes Cole a fit for a build-in-public angle, a webinar on an open-source factory and lessons from operating one, or a guide for teams seeking an out-of-the-box starting point rather than a blank architecture. Cole Medin also states that an open-source AI software factory is a priority, giving the campaign a concrete artifact for a demo or workshop.
- Bartosz Jurga: Choose Bartosz Jurga when the developer-tools company wants to lead with security, governance, or board-level risk inside software factories. The posts cover a package compromise and a campaign that infected more than 400 packages with more than 500 million weekly downloads. They argue that perimeter defense is irrelevant if attackers enter through trusted delivery pipelines. They address CEOs, CISOs, and CTOs on blast radius, deployment secrets, compliance exposure under NIS2 and ISO 27001, business continuity, and ransomware decisions. Recommendations include version pinning, least-privilege pipeline access, automated credential rotation, immutable backups, air-gapped storage, and tested recovery. This is a stronger alternative when the technical webinar or guide should focus on securing the factory and recovering from supply-chain attacks, rather than configuring agents and runners.
- Darius Radford: Choose Darius Radford when the sponsor wants a named operating model and assessment offer around secure software factories, rather than the factory-building stack. The posts describe a Secure Software Factory framework with four capability pillars — orchestrated delivery, developer-centric application security, artifact and supply chain control, and unified risk correlation — and a 2.0 version that adds post-deployment protection, with AI governance across the model. They also reference WeAreDevelopers World Congress North America sessions on the agentic software factory and explain why faster pipelines and AI-generated code increase the need to verify dependencies, enforce policy, understand risk, and protect production. The assessment path examines how an organization builds, tests, packages, deploys, and operates software, then sets a maturity baseline and roadmap. It suits a webinar or guide for security and platform leaders seeking a structured way to start.

---
AI-assisted summary of public posts. Audience demographics are not available. Engagement appears only where we checked recent posts. Method: https://b2binfluencer.marketing/methodology/