June 2026 Summaries

Factory’s AutoWiki is a documentation-generation tool that analyzes source repositories to create structured, browsable, and continuously updated wikis, positioning documentation as a build artifact rather than a separate maintenance task. It uses a multi-agent process to survey codebases, plan wiki structures, generate pages and diagrams, capture visuals, and optionally produce narrated walkthrough videos, covering areas such as architecture, technology stacks, project structure, entry points, systems, and contribution guidance. AutoWiki can refresh documentation manually through the Droid CLI or automatically on every push through GitHub Actions or GitLab CI, using incremental generation to update only pages affected by code changes. Generated wikis are available through Factory’s web viewer with search, version history, batch refreshes, and videos; synced to GitHub wiki tabs; accessible in Droid sessions; and committed within repositories under a droid-wiki directory. Available on all Factory plans, the tool is presented as a way to speed onboarding, help engineers and AI agents navigate unfamiliar codebases, and keep repository documentation aligned with evolving software.

Factory natively hosts and serves MiniMax open-weight language models through its Droid Core offering, allowing users to run them without supplying inference infrastructure or API keys. Available variants include MiniMax M3, M2.7, and M2.5, which can be selected in interactive Droid sessions through the model menu or specified in headless commands, CI workflows, and Missions using their model IDs. M3 and M2.7 use High reasoning, while M2.5 supports Low, Medium, and High levels. All listed MiniMax models have a 0.12x credit multiplier, currently the lowest in Factory’s catalog, making them suited to high-volume, cost-sensitive tasks such as scoped code changes, cleanup, experimentation, and automated workloads. Users can assign MiniMax to worker agents while reserving stronger models for validation, switch models by task, or use Factory Router to choose models automatically.

Factory 2.0 presents a vision of “software factories,” interconnected AI-native systems that manage the full software development lifecycle, from external signals such as bug reports and customer feedback through planning, coding, testing, security review, deployment, monitoring, and continuous feedback. The company argues that these systems should support model independence, allowing organizations to select AI models based on cost, speed, and performance; sovereign intelligence, enabling control over hosting, data, and organizational learning; and continual self-improvement through shared context across development processes. Factory describes autonomy as a gradual, organization-specific progression, ranging from simple agents and workflow automations to persistent execution and multi-agent “Missions” for complex work. It says software factories are already being deployed by several large organizations and announces expanded management visibility in its desktop application, while emphasizing that engineers will increasingly be responsible for designing, governing, and operating the systems that build software.

Factory has introduced automated security review in Droid, which performs STRIDE-based assessments on every non-draft pull request alongside standard code review and posts validated findings as inline comments with severity levels, CWE references, explanations, and suggested fixes. The system targets threats including spoofing, tampering, information disclosure, denial of service, privilege escalation, OWASP Top 10 and LLM Top 10 issues, injection, authentication weaknesses, and secrets exposed in logs. Droid creates a lightweight threat model of changed code, scans for vulnerabilities, validates potential issues against the diff to reduce false positives, and provides a deduplicated PR summary. The company cites production audits that led to responsibly disclosed findings, including CVE-2026-42876 in external-secrets and a webhook signature verification issue in the WorkOS Node SDK. Users can install the feature through Droid’s code-review setup, run on-demand repository or diff audits with a CLI command, and use deeper multi-agent full-repository audits through GitHub Actions or Droid Missions; the capability is available on all plans.

Factory is expanding its presence to Sydney to better serve its expanding customer base in Australia and the Asia-Pacific region. The move aims to leverage Australia's renowned engineering talent in sectors such as finance, technology, telecommunications, mining, energy, logistics, and the public sector, where organizations are increasingly adopting autonomous Software Factories. These Software Factories operate continuously with human oversight, enhancing productivity and innovation. Establishing a hub in Sydney allows Factory to collaborate more closely with local customers and partners. The company is actively recruiting across various functions and invites interested candidates to contact them via [email protected].

Factory, a company specializing in enterprise AI coding agents known as Droids, has appointed Marcello Gallo as its Chief Revenue Officer, following his successful tenures at Sigma and Moveworks where he significantly increased revenue growth. Gallo's strategic leadership is expected to guide Factory through a phase of rapid expansion, as enterprises increasingly adopt agent-native strategies beyond initial experiments. Factory, which facilitates the deployment of AI agents across the software development lifecycle for clients like NVIDIA and Morgan Stanley, recently secured $150 million in Series C funding, valuing the company at $1.5 billion. It also launched Factory Router, a product designed to optimize coding tasks by selecting the most suitable model, thereby reducing costs while maintaining top performance. Factory's platform aims to revolutionize enterprise technology by allowing companies to build autonomous software systems with human oversight, enhancing flexibility and control over enterprise code.

Factory natively hosts and serves DeepSeek V4 Pro, an open-weight DeepSeek model available through its Droid Core catalog, allowing users to use it without managing inference infrastructure, GPUs, or provider API keys. Users can select the model in interactive Droid sessions with the model menu or keyboard shortcut, or specify its `deepseek-v4-pro` ID for headless execution, CI workflows, and Missions worker or validator agents. DeepSeek V4 Pro supports Off, Low, High, and Max reasoning levels, with High as the default, enabling users to balance speed and deeper problem-solving needs. Factory bills the model at a published 0.7x credit multiplier, though users are advised to check current pricing before adopting it broadly. The platform also supports switching models by task or using Factory Router to automatically choose models across its catalog.

Factory Router is a tool designed to optimize the cost and performance of AI models by automatically selecting the most suitable model for each task, thereby reducing token spend by 20-25% while maintaining high performance levels. It addresses the challenges faced by engineers who often default to using the most performant, yet costly, models for tasks that do not require such high capabilities. By routing tasks to the appropriate models based on their requirements, Factory Router ensures efficiency and reliability, even in situations where providers degrade or face capacity constraints. It provides enterprise-grade reliability with 99.9%+ request reliability by dynamically selecting paths across different models and providers. The tool also allows for customizable routing guidance to reflect an organization's specific workflow needs, offering standard controls over model availability and governance. Currently in private research preview, Factory Router aims to make high-quality, cost-effective operations the default for organizations by effectively balancing the cost/performance trade-off, as illustrated by its performance on benchmarks like Terminal-Bench 2 and Legacy-Bench relative to Claude Opus 4.7.