Tell us where it breaksContact
MenuClose

Pangalom · App studio & AI agents

Most AI agents never leave the demo.

We build the ones that reach production. Scoped, observable, reversible. An app studio: iOS, Android and web, the enterprise systems underneath them, and the agents that do the work.

Tell us where it breaksSee how we work

An agent that cannot be bounded, watched and undone is not a product. It is a liability with a chat interface.

Why we exist

The work will not disappear. It will be divided.

Every organisation is about to draw a line through its own operations. On one side, the work a machine does. On the other, the judgement a person keeps. Most will draw that line by accident, at whatever depth the first tool they bought happened to reach.

The line matters more than the model. Put it in the wrong place and a company loses the thing that made it good at its job: knowledge that was never written down, that lives in the people doing the work, and that no vendor can hand back once it is gone.

Vision
Organisations that draw that line on purpose, and come out stronger on both sides of it.
Mission
To draw it with them, and to build the systems that hold it.

Why we do research and delivery

That line cannot be drawn from a laboratory, because what decides where it belongs is knowledge held inside a business rather than inside a model. It cannot be drawn from a consultancy either, because holding it means building things that do not exist yet. So we do both. We research how these systems should behave, and we ship the ones our clients run.

What we do

Four things, engineered properly.

Apps

iOS, Android and web, designed as well as built. Published from your own developer account under your own name, with the source in your own repository from the first day.

  • App Store and Play
  • Design and build
  • Published in your name

AI agents

Agents built against one operation, trained on how that organisation actually works. Each one runs inside a written authority and stops at its edge.

  • Scoped authority
  • Human approval gates
  • Full action audit

Enterprise systems

The software underneath: the workflows, the records and the reporting that a business currently holds together with spreadsheets and standing meetings.

  • Workflow engines
  • Reporting surfaces
  • Migration off spreadsheets

Integration

Agents are only useful where the data already lives. We connect to the systems you run today rather than asking you to replace them first.

  • ERP and finance stacks
  • Ticketing and CRM
  • Internal APIs

In practice that is a customer app on iOS and Android, an internal tool that replaces the spreadsheet a department is running on, an agent that works a queue and escalates what it cannot decide, and the connection between two systems that were bought separately and were never going to talk to each other.

We are building our own products on the same engineering. They are not ready, so they are not described here, and we will name them when they ship rather than before.

The agent problem

The hard part is not the model. It is the boundary.

Running an agent is easy. Three questions are not: what is it permitted to do, what happens when it is wrong, and who is accountable for the outcome. Demos skip all three. Production does not.

Scoped
The agent operates inside a defined authority and cannot act outside it. The boundary is written down and agreed before anything is built.
Observable
Every action is logged, attributable and explainable. If nobody can answer why it did that, it is not finished.
Reversible
Nothing irreversible happens without a person approving it. The default is to propose, not to execute.

And we do not fit a general model over a business. We train the agent on how that organisation actually works.

If it does not return a measurable hour or a measurable cost, it does not ship.

Every engagement

The system is yours. These six are not negotiable.

We do not resell the last client's system to the next one, because an operation that works like everybody else's is not worth automating. What does not change is the machinery around the agent. The model is the smallest part of the work; this is most of it, and it is the same whatever we build.

  • Authority model

    What the agent may do, written as policy rather than buried in a prompt.

  • Approval gates

    Irreversible actions queue for a named human. Nothing slips through by default.

  • Audit trail

    Every decision recorded with its inputs, so an outcome can be reconstructed months later.

  • Evaluation

    Agents are measured against real cases before they touch production, and after.

  • Failure behaviour

    When the agent cannot proceed it says so and stops. It does not improvise.

  • Handover

    Runbooks, escalation paths and a team that knows how the system behaves under load.

This is built for organisations where a wrong action is expensive: where an auditor will ask who approved something, where a process has more exceptions than rules, and where the people who understand the work are the ones you cannot afford to lose.

Built on

The tools underneath the work. We build on infrastructure already proven at scale, because an agent is only ever as dependable as the system running it.

  • Anthropic
  • OpenAI
  • NVIDIA
  • AWS
  • Cloudflare
  • Kubernetes
  • Terraform
  • Temporal
  • PostgreSQL
  • Kafka
  • Redis
  • OpenTelemetry
How we engage

Understand, build, operate.

  1. 01

    Understand

    Operations map, costed shortlist

    We embed with the people doing the work and map how it really happens, including the parts that never made it into the process document. You keep the map whether or not you build with us.

    Deliverable: an operations map and a costed shortlist

  2. 02

    Build

    Working software, in your environment

    We ship in increments you can evaluate, against a boundary agreed in advance. No twelve-month programme with a reveal at the end.

    Deliverable: working software, in your environment

  3. 03

    Operate

    Handover, runbooks, support line

    Your team runs the system and we stay accountable for it. The people who used to do the work now review what the system proposes.

    Deliverable: handover, runbooks and a support line

Beyond the Possible

You do not have to believe any of this.

There is no client list on this page and no numbers on it you cannot check. What there is instead is a way to find out, on your own work, at a price you can walk away from, before anything is built.

  • Three weeks

    A fixed fee, an operations map and a costed shortlist. Yours whether or not you build with us, and yours to take to somebody else.

  • Your own cases

    Nothing goes near live work until it has been run against what already happened in your business, and you see how it did before you decide anything.

  • One increment at a time

    The build arrives in pieces you can evaluate separately, priced separately. You can stop after any one of them and keep what you have paid for.

Contact

hello@pangalom.com

Tell us what the work looks like today and where it breaks. If we are not the right people for it, we will say so.

Pangalom on LinkedIn