Systems of action for humans and agents: Tables, actions and a record of every change

A system of record tells you what is true: the customer, the invoice, the policy. A system of action does something about it. When a new row arrives, it checks the row, enriches it, splits it, sends it on or turns it into a file — and it keeps a record of what it did and why.

That is the whole idea behind Morphcloud, and it rests on three parts.

Tables your team already understands

Everything in Morphcloud starts in a table. Columns are typed — text, numbers, currency, percentages, dates, files and more — and tables link to each other through references, so an invoice knows its client and a ledger line knows its invoice.

Rules live in tables too. If an events company splits every conference invoice 35% to staffing, 30% to venue, 20% to kit and 15% to travel, those percentages are four rows in a table. Change a row, and every new invoice follows it.

Actions that run on every row

An action is a step that runs on a row. Morphcloud has nine kinds:

• Agent — one AI model call per row, for classifying, extracting or summarising.

• Web — call any HTTP API and bring the answer back into the row.

• Script — a short piece of JavaScript for exact rules and arithmetic.

• Lookup — read or total values from another table.

• Network — copy a row into another table, which starts that table’s own actions.

• Join — turn one row into many, such as one invoice into one line per cost centre.

• Email — send a templated email from a connected mailbox.

• Generate — produce an Excel, CSV or Word file, including the recipient’s own Excel template.

• Load — import a spreadsheet through a queue where you fix headers, dates and bad values first.

You write an action once and connect it to any table. On each table you set the order the actions run in, conditions that decide which rows each action runs on, and timers that run work on a schedule, from every minute to weekly. You can also start from more than 100 ready-made actions in the public library, from geocoding and VAT checks to web scraping and document conversion.

A record of every change

Work you can’t see is work you can’t trust. In Morphcloud:

• Every cell keeps its history — each earlier value, when it changed, and whether a person or a named action changed it.

• Every action keeps a run log — inputs, outputs, errors and timing for each run.

• Every row shows the status of its work, so you can see which rows are still being processed.

• The workspace graph shows how every table and action connects, and flags automations that could trigger themselves in a loop.

When something doesn’t add up, the answer should be on the row, in words — not buried in a log only an engineer can read.

For humans and for agents

People work in Morphcloud through tables and forms they already understand. AI agents work in it through an API that describes itself and an MCP server, so an assistant such as Claude can create tables, connect actions, test them and fix them, using your account and your permissions.

We tested this ourselves. Our Invoice Cost Allocation showcase — four tables, six actions, a reconciliation check and an Excel export in the finance team’s own template — was built by an AI agent working over MCP, in one session, with a person reviewing each step. Each workspace can also hold a written brief — the goal, how you’ll know it’s done, and what has been proven so far — so people and agents work from the same plan.

What it costs

You pay for the data you store and for the work that changes it. Table data and documents are billed per GB per month. An action run costs one credit — a tenth of a cent — and a run that changes nothing costs nothing. AI models are billed per token at published rates.

What comes next

Over the coming weeks we’ll publish the use cases behind this: splitting and reconciling invoices, turning incoming spreadsheets into checked records, and collecting figures from public websites. Each one is a live workspace, and each post shows how it works — including the parts that don’t add up.

If there’s a spreadsheet your team retypes every week, book a demo and bring it with you.