Custom software, built and run by AI agents
Describe what you need in a conversation. Software Factory writes the spec, builds the change, tests it, and ships it to production. People step in when a decision needs judgment.
For companies of tens to hundreds of people that run on spreadsheets, group chats and too many SaaS tools.
When a salesperson offers more than a 15% discount, I want the regional manager to approve it. Also, salespeople shouldn't see our internal cost.
Should the 15% threshold be calculated from list price or from the salesperson's assigned price level?
From list price.
Software Factory
The discount approval workflow is live. Discounts above 15% now need regional manager approval, and internal cost stays hidden from sales reps.
The problem
A manufacturer changes its purchasing steps to fit an ERP. A clinic types the same patient details into three systems. An online store keeps supplier exceptions in spreadsheets because its order system can't express them.
What changes
Writing code was never the only slow part. Requests wait between account managers, product managers, sprints, reviews and releases. AI agents can now carry a change through those steps without the queues.
Planning, review and testing still happen. They run inside the factory.
What we measure
The time from a confirmed need to safe, running production software. For ordinary business changes, the target is hours or days.
How it works
Every request moves through the same stations. You take part in the first one. The factory runs the rest and reports back.
You describe the need in WeCom, Slack, Teams, email or web chat. The product agent already knows your processes and asks only questions that change the outcome.
The request becomes requirements, acceptance criteria, dependencies, and a test and rollout plan.
Coding agents implement the change on tested foundations. The control plane routes each task by type, risk and cost.
Unit, integration, browser and permission tests collect evidence that the business behavior works.
The change ships under the review policy that fits its risk, with smoke tests and a rollback path.
The factory watches errors, performance and workflow completion after release. Problems become new work.
Then the line starts again with your next request.
Example
You supply the business intent and settle real ambiguity. The factory writes the rest.
Our sales managers should approve discounts above 15%.
Verification evidence
Tests check the business rule, not only whether a screen renders.
Work item
People and agents
Escalation depends on uncertainty and consequence, not technical difficulty. A large refactor with strong tests can run on its own. A one-line change that deletes financial records waits for a person.
| Situation | How it's handled |
|---|---|
| Well-defined change with low consequences | Built and deployed automatically |
| Technically complex change with low consequences | Automatic, with stronger testing |
| Ambiguous business decision | The product agent asks you |
| Major architecture decision | Human review |
| Destructive database migration | Explicit approval |
| Authentication, billing, financial, security or permission change | Elevated review rules |
| Production incident | Predefined mitigation, escalated when unusual |
Autonomy grows as the factory builds a record of reliable delivery. At first, every production deploy needs a person to approve it.
Traceability
The factory keeps a product model for each customer. It links conversations and decisions to code, tests and deployments, so later agents can tell deliberate behavior from accidents.
Why does this field exist?
It was added because the Shanghai finance team needed to tell reimbursement entities apart.
Who it's for
Good candidates depend on several SaaS tools, lean on spreadsheets and spend hours coordinating work by hand.
Quotes, purchasing, production planning, quality and shipping often outgrow the ERP's standard workflows.
Improve intake, scheduling, staffing and follow-ups around the clinical systems you already use.
Connect orders, suppliers, inventory, pricing and returns across Amazon, Shopify, Taobao, JD, PDD and TikTok Shop.
Also distributors, logistics companies, professional service firms and multi-location businesses.
Foundations
A shared platform covers what every business application needs. Business modules are then selected, combined and adapted to your workflow.
CRM
Contacts, companies, opportunities, pipelines
Inventory
Products, warehouses, stock movements, reservations
Purchasing
Suppliers, purchase orders, receiving
Sales
Quotes, orders, discounts, fulfillment
Service operations
Appointments, staff, schedules, service records
Workflow
Approvals, tasks, notifications, escalations
E-commerce operations
Orders, returns, campaigns, marketplace integrations
Your workflow
When no module fits, agents adapt one or build a new one.
Identity and access
Sign-in, organizations, teams, roles and permissions
Application interfaces
Forms, tables, search, dashboards, reports and languages
Collaboration and workflow
Comments, attachments, notifications, email, approvals and background jobs
Data and integration
Imports, exports, APIs and webhooks
Delivery and operations
Tests, audit logs, monitoring, backups, feature flags and preview environments
Continuous evolution
Traditional custom software ends at delivery, and every later change is a new engagement. With Software Factory, a request like this one is ordinary operating work.
Starting next month, purchase orders above $50,000 need CFO approval.
Initial implementation
Process analysis, data migration, integrations and setup
Platform subscription
Hosting, monitoring, backups, security updates and the product agent
Continuous development
New features, automations and support with agreed response times
Premium requirements
Dedicated environments, compliance needs and advanced integrations
The factory works out whether that needs an integration, an automation, a new screen or a new application. It builds and checks the fix, ships it, and watches whether the problem went away.
Our warehouse team wastes two hours every day copying shipping information from WeCom into the ERP.