SWFT is a free publication from Hraness about software factories, the repeatable systems where AI agents do scoped work while automated checks and people decide what ships. Its company cases, including Stripe, Shopify, Jane Street, and Capital One, show how work reaches the agents, what checks it, and where a person decides, and one table compares every case on the same fields. A scale reference compares reported agent numbers and keeps each figure beside its unit and limits. Every article links its sources.
Who is SWFT for?
SWFT is for product and business leaders who want to understand how an agent system works, judge a claim about its scale, and find relevant examples. The guides connect engineering concepts to decisions about customers, tradeoffs and delivery.
Engineers and founders are welcome too. SWFT uses the accepted AI engineering terms so readers can join the wider conversation, then translates each term into everyday language and a concrete product decision.
What does SWFT publish?
- Ideas compare the arguments changing software development. That includes Geoffrey Huntley's single outer Ralph loop, Steve Yegge's multi-agent work graph, whether the coding-agent harness is a cost lever, and the counterarguments about expertise, context, ownership, and maintainability.
- Company cases explain how a named organization moves work from request to reviewed result, which patterns transfer, and what remains unproven.
- Across industries compares documented use in software delivery, business operations and research. It separates systems in daily use from pilots and product capabilities.
- Scale records public measurements of agent teams, runs, workload, output and adoption. Every number keeps its unit, time window, outcome and source.
- Guides answer practical questions about architecture, pilots, evals, quality, economics, human attention, and release.
- References keep definitions, comparisons, and the field map current for people and machines.
What is SWFT's point of view?
There is no single software-factory shape. A company might automate one repeated task, give an agent a loop for checking and improving its work, or coordinate a team of agents through a shared task system. These approaches can work together. None is automatically better because it uses more agents. What matters is the job, the information and tools available, the checks on the result, and the people responsible for it. Human attention is often the limiting resource.
Scale is part of the story. A system that can coordinate thousands of workers opens different possibilities from one person delegating one task. SWFT tracks those possibilities alongside the cost of coordination, the way results are checked, and the people involved. It also follows widespread use of smaller systems, which can matter across an industry even when no single run is enormous.
Company accounts remain attributed. Studies keep their scope and limits. A count of publicly documented cases establishes where activity is visible; it does not estimate the percentage of an industry using agents. Historical observations remain dated so a useful reference does not depend on each example being the newest or largest.
Start with the industry overview to see where agent systems are appearing, or the scale reference to inspect the numbers. Three visions of the AI software factory explains the field's competing approaches. The coding-agent harness is a cost lever reads HarnessTax for factory economics. The rise of software factories is Ben Guo's personal manifesto. The methods page explains sources, measurements and corrections.
How is SWFT different from other sources?
Vendor pages such as Port's and Factory's explain software factories through their own products. ZenML's LLMOps Database and Evidently AI's case-study list collect uses of AI and machine learning across organizations.
SWFT focuses on named companies' own agent systems, most of them for building and running software. The company comparison puts every case on the same fields: main job, reported use, human checkpoint and access. Each page distinguishes observed artifacts, independent studies, first-party reports and SWFT's analysis.
Who writes SWFT?
SWFT Editorial is the institutional byline for the source-led guides, comparisons, references, and company cases. Those pages are AI-drafted from their cited public sources. A second AI agent, Codex or Devin as each page states, then checks that each claim stays within what its source supports, that the page does not repeat another SWFT page, and that it is useful to the reader. The institutional byline does not imply that Ben Guo personally wrote or reviewed them.
The rise of software factories is different. Ben Guo wrote it and first published it on Substack. SWFT republishes it with the original body, links, date, byline, and sign-off. It is the only page under Ben's name.
The publication is editorially independent. Inclusion in the map is not an endorsement, and coverage is not offered in exchange for payment.