What Happens When Your AI Agent Runs Code for You

Most AI tools are chatbots. You type a question, you get a text answer. Useful for brainstorming, maybe drafting an email. But the moment you need to actually do something with data – crunch numbers, transform a spreadsheet, pull figures from an API – a chatbot hits its limits.

Sprigr agents are different. They don’t just talk about your data. They write code, execute it in a secure sandbox, and hand you the result. All from a plain-English conversation.

This guide explains what that means in practice, how the sandbox works, and what you can actually do with it.

Beyond chatbots: agents that execute code

A chatbot generates text. An agent takes action. The difference matters when you’re running a business.

Ask a chatbot to “calculate the average job value from this CSV” and it might give you a rough estimate or tell you how to do it in Excel. Ask a Sprigr agent the same thing and it writes a Python script, runs it against your actual data, and returns the precise answer – along with a breakdown by job type if that’s useful.

This is code execution, not code suggestion. The agent writes the script, runs it, reads the output, and uses the result to continue the conversation or feed into the next step of a workflow.

How the sandbox works

When a Sprigr agent needs to run code, it executes inside an isolated sandbox environment. There are two runtimes available depending on the task:

• Python (container-based). Full Python environment with access to popular libraries like pandas, requests, and openpyxl. Execution typically takes 2–5 seconds. Ideal for data processing, file transformations, API calls, and report generation.
• JavaScript (V8 isolate). Lightweight execution in under 100 milliseconds. Perfect for quick calculations, JSON transformations, and string processing where speed matters.

The agent chooses the right runtime automatically based on what it needs to accomplish. You don’t need to specify a language or configure anything.

Auto-injected credentials

Here’s where it gets powerful. When you connect integrations to Sprigr – simPRO, Xero, Gmail, or any other service – those API credentials are automatically available inside the sandbox at runtime.

That means the agent can write a script that calls the Xero API to pull your revenue figures, and it doesn’t need to ask you for credentials. They’re already there, securely injected into the execution environment. You never paste API keys into a chat. You never store them in a script. The agent simply has access to the same integrations you’ve already authorised.

Credentials are encrypted at rest and only decrypted at the moment of execution. They’re scoped to your workspace and never shared across tenants.

Real examples: what you can actually do

Process a CSV of supplier prices

Upload a supplier price list and ask the agent to compare it against your current rates. It writes a script that reads both files, matches items by SKU or description, calculates the differences, and outputs a clean summary showing where you’re overpaying.

Generate a monthly revenue report from Xero

Ask “give me a revenue breakdown by customer for March” and the agent calls the Xero API, aggregates invoice data by contact, and returns a formatted table. No spreadsheet wrangling. No manual exports.

Transform simPRO job data into a custom format

Need to send job summaries to a client in a specific format they require? The agent pulls job data from simPRO, restructures it to match the client’s template, and generates a downloadable file. What used to take an hour of copy-paste takes seconds.

Build a one-off data migration script

Moving from one system to another and need to reformat hundreds of records? Describe the source format and the target format in plain English. The agent writes a migration script, runs it against your data, and gives you the transformed output ready to import.

Calculate pricing and margins

Upload your cost sheet and markup rules. The agent builds a pricing calculator that applies your margin structure, accounts for material surcharges, and produces a quote-ready price list. Update the inputs any time and re-run.

The workspace file system

Each Sprigr workspace includes a persistent file system that the sandbox can read from and write to. This means:

• Upload files for the agent to process – CSVs, spreadsheets, JSON exports, text files.
• Download results that the agent generates – reports, transformed data, formatted documents.
• Persist data between conversations. Files created in one session are available in the next. Store reference data like price lists, customer mappings, or templates that the agent can use repeatedly.

The file system is scoped to your workspace. Other tenants cannot access your files, and the agent only reads files you’ve uploaded or that it has created on your behalf.

Security and isolation

Letting an AI agent run code raises obvious questions. Here’s how Sprigr handles them:

• Isolated execution. Every code run happens in its own sandbox. There is no shared state between executions unless you explicitly persist files to the workspace.
• No cross-tenant access. Your sandbox cannot reach another customer’s data, credentials, or files. Tenants are fully isolated at the infrastructure level.
• Audit logged. Every piece of code the agent writes and executes is recorded. You can review exactly what ran, what data it accessed, and what output it produced.
• Scoped permissions. The agent only has access to the integrations you’ve connected. It cannot reach services you haven’t authorised.
• No outbound network by default. Sandbox execution can be restricted to only approved API endpoints, preventing data from being sent to unexpected destinations.

When to use code vs built-in tools

Sprigr agents have built-in tools for common tasks – creating simPRO jobs, sending emails, looking up customer records. These are faster and don’t require code execution at all.

Code execution is for everything else. The situations where no pre-built tool exists for what you need. Specifically:

• Use built-in tools when the task is standard – create a job, send an email, update a record, look up a contact. These are optimised, reliable, and instant.
• Use code execution when you need custom logic – data transformations, calculations across multiple sources, report generation, file format conversions, or anything that requires looping through records with business-specific rules.

The agent handles this decision automatically. It reaches for built-in tools first and drops into code execution when the task requires it.

Getting started

Code execution is available in every Sprigr Team workspace. There’s nothing to enable or configure. Connect your integrations, start a conversation, and ask the agent to do something that requires real computation. It handles the rest.

Start simple. Upload a CSV and ask the agent to summarise it. Ask it to pull your last month’s invoices from Xero and calculate your top customers by revenue. Once you see how it works, you’ll find dozens of tasks that used to require spreadsheet gymnastics or a developer’s time.