AI software factory: senior engineers + agents.
Pento runs every project on Steer, the platform we built: from ERP workflows and legacy migrations to AI features, data pipelines and integrations. Senior engineers write the spec, agents write the code on branches in your GitHub, and merges follow the policy you set: low-risk changes ship on a green review and CI, and a person approves the rest.
Every project starts from a written spec, split into tasks with acceptance criteria. Each task comes back with an evidence report of what changed, the tests that ran and screenshots of anything on screen.
Every cloud run gets a fresh, isolated machine that is thrown away when the run ends. Each git action uses a short-lived token that reaches only your project's repositories, and merges follow the policy you set: low-risk changes ship on a green review and CI, and a person approves the rest.
Steer breaks each spec into ordered waves of tasks. Several agents work at the same time, each in its own machine, while our engineers review what comes back.
What is an AI software factory?
An AI software factory is a delivery model where senior engineers write the spec, AI agents build in parallel in isolated environments, and every change ships with evidence and the approvals its policy requires. Pento runs its AI software factory on Steer, the platform we built, and delivers each project at a fixed price and date.
Coding agents cut the cost of building software. The spec and the review decide whether it is good.
Most companies carry software they rent or keep putting off: the workflows the ERP won't do, the migration that never made the budget, the report someone rebuilds by hand every month. Agents can now write much of that code.
Left unchecked, they also ship bugs and security holes. We built Steer around that risk: every task starts from a written spec, every cloud run happens in a fresh machine that is thrown away afterward, and each finished task comes back with an evidence report your reviewers read before anything merges. That is also how we can quote a fixed price and a fixed date.
Agents do the implementation. Our engineers spend their time writing the spec and checking the result.
Steer is built by Pento. Our engineers have shipped production software for more than 15 years, including AI search that runs for clients like Mercado Libre. We run Steer on ourselves too: we are replacing our own ERP with it, module by module.
Every company has a system it has learned to work around.
The ERP keeps the books while quoting, scheduling and approvals live in spreadsheets around it. Vendors bill monthly and ship their own roadmap. Agencies quote in hours, and the date moves. We build the piece you are missing one bounded project at a time, each quoted with a fixed price and a fixed date, and the code lives in your repositories.
The workflows NetSuite or SAP won't do
Scheduling, quoting, customer portals and approvals, built as satellite apps around your system of record, which keeps the books as it does today. Each piece is specced, checked and approved on its own, and integrated with the ERP instead of fighting it.
Describe the workflow →The migration that never made the budget
The vendor got acquired, the codebase predates the cloud, and every idea gets filed under “after the migration.” Those quotes were rational when humans typed every line. We move one subsystem at a time, each piece specced, verified and approved before the next one starts, so you never bet the year on a rewrite.
Plan the first subsystem →When native tools stop being enough
AI search and recommendations tuned to your catalog, copilots and automation wired into the product you already have, and agent workflows (MCP servers, skill packages, multi-agent systems) for teams pushing into agentic territory.
See AI workflows →✳ The same approach covers document management, forecasting with ML, data engineering and custom integrations. A first conversation shows where to start.
Every kind of custom build. Only faster.
If it is software, we ship it: AI in the loop where it earns its keep, senior engineers in the loop where judgment matters. Each of these is a project with its own scope, price and date, running on Steer through the same evidence and approval steps.
ERP and operations
Scheduling, quoting, portals and approvals that NetSuite or SAP won't do, built as satellite apps you own, connected to your system of record.
Legacy modernization
Aging systems moved one subsystem at a time, each piece specced, checked and approved before the next one starts.
AI features and integrations
Agents, copilots and automation wired into the product you already have, plus MCP servers, skill packages and multi-agent workflows.
Ecommerce and search
AI search and recommendations tuned to your catalog, and agent-ready storefronts so shopping agents can find, compare and buy from you.
Internal platforms
Admin panels, dashboards and connectors: the operational backbone that quietly runs the business.
POCs and MVPs
Validate an idea before you commit a quarter to it, then get a real product in front of real users.
A workspace your team can see inside
Most AI development happens in a black box. Steer runs in the open: goals and specs come first, agents build in parallel sessions, and every pull request carries its review trail of tests, screenshots, and verdicts. Your team has a seat alongside ours from day one, so progress is something you watch, not something you are told about.
How the AI SDLC works at Pento
Define goals together
We sit with your team and turn the roadmap into structured goals with clear acceptance criteria. Product intent, constraints, and existing context (docs, designs, repos) get captured once and kept in the project's context store.
Spec and plan
The orchestrator and our engineers decompose each goal into specs and sized pieces of work. Scope is fixed when the price is: you approve a plan with a fixed date before any code is written.
Agents build in sandboxes
Coding agents execute in isolated environments, each with a brief, the relevant context, a budget, and a time limit. Senior engineers steer them in parallel, so a project that used to take a quarter ships in weeks.
Review with evidence
Every change set is judged by an independent review agent and a senior engineer: spec conformance, standards, security, tests. Verdicts, findings, screenshots, and traces become evidence attached to the PR.
Ship, measure, learn
Merges, releases, and telemetry (lead time, cost, rework, adoption) are tracked in the same workspace, and decisions and lessons feed back into the project's memory.
We don't sell our platform. We ship with it, every day.
Steer is how our engineers work, sharpened on every real project, internal and external. When you engage us, you get the same tooling and the same loop we use to ship our own systems.
Case studies →Our own ERP, replaced module by module
We run Steer on ourselves first: invoicing, projects and allocation moved off our old system one bounded piece at a time, each specced, evidenced and approved before the next.
A 2-week migration off a 2018 monolith
Subsystem by subsystem instead of a big-bang rewrite. Zero post-handoff incidents.
An 8-day MCP integration into a client's CRM
A custom agent workflow wired into systems that were never designed for agents, with end-to-end behavioral tests included.
A 3-week internal admin tool, ~120 PRs
Agents executing in parallel, engineers reviewing evidence on every pull request. Shipped on the quoted date.
Software factory as a service, not another tool
Nothing is quoted before the audit. The same engineers run all three steps in the same workspace.
Audit
We review the systems you run, the work people do by hand around them, the data moving between them and the software you rent. You get a written report of what we found, and it is yours to keep whatever you decide.
Consulting
We read the findings next to your business goals and your numbers to pick the quick wins and the projects with the highest return. You leave with a one-page scope, a fixed price and a fixed date for the first one, before any code exists.
Execution
Our engineers and AI agents deliver the project on Steer, in a workspace your team signs in to. Each finished task comes with an evidence report, and merges follow the policy you set: low-risk changes ship on a green review and CI, and a person approves the rest. After launch, the same engineers can keep it running.
Why build with Pento and Steer
We built the methodology and the platform ourselves, and we run our own company on them: we are replacing our own ERP with Steer, module by module. When you engage us, you get the same loop and the same tooling we trust with our own business.
What every Steer engagement includes:
The questions that come up on every first call
Contact us
Let's build your next project on Steer
Tell us what your roadmap needs: a bounded project or an ongoing build. We will walk you through the platform and the fastest path to shipping. An engineer reads every request.