What is an AI agent, in practical terms?
An AI agent is software that understands requests written in normal language and takes predefined actions in your systems: it looks up an order, creates a lead, drafts a reply, routes a ticket or updates a record. It differs from a scripted chatbot because it copes with messy, free-text input, and from a plain AI chatbot because it can act. See agent vs chatbot.
Which agents do we build?
- Enquiry and lead agents that qualify WhatsApp or website enquiries and create CRM records
- Support triage agents that classify, prioritise and draft answers from approved content
- Order and operations agents that read free-text orders and prepare drafts for approval
- Reporting agents that pull data from several tools into a daily summary in chat
- Watch agents that check prices, ads and site health and raise alerts
- Research agents that gather public information into structured workbooks with sources
How do we know these agents work in real life?
We run them ourselves. The owner's own businesses use a 14-bot SEO team, a Google and Meta Ads agent, a website ops agent, watch routines, a WhatsApp manager, social media and brand visuals agents. They're in-house systems, not client results, but they've taught us where agents break, what needs approval and how to report so a person stays informed. See the agents catalogue.
What makes our agents different?
- A written stop rule for every agent, covering what it does alone, what needs approval, hard stops and the handover package.
- Approval for money and irreversible actions: spend, refunds, deletions and published claims.
- Backups and change logs before any edit to a live system.
- Daily reports in chat, so you see what the agent did without logging into a dashboard.
- Narrow agents over one giant agent: several focused agents are easier to test, trust and fix.
Which systems can an agent connect to?
WhatsApp Business Platform, CRMs (Zoho, HubSpot and others), helpdesks, Google Sheets and Excel, Shopify, WordPress, Google Ads and Meta Ads, accounting exports, Supabase and Postgres, and most tools with an API. We choose the AI model and plumbing (for example n8n) per workflow. See tool comparison.
What does an agent pilot deliver?
One agent for one workflow, with a written scope and stop rule, built into your tools, tested on real cases, reviewed with your team, and documented so it isn't a black box. Voice agents are available to build as a pilot, but we don't run one live today.
The agent drafts, routes and escalates. A person decides. Before anything is built we write down what the agent may do alone, what needs approval, and when it must stop and hand over with the full context attached.
How is the agent team put together?
Every team we build uses the same architecture: a team lead agent that plans and assigns work, specialist agents with narrow jobs and permissions, a shared task board where every task and approval is visible, tools and connectors limited to what each agent needs, scheduled routines for recurring checks and reports, and approval gates wherever money, customers, data or irreversible changes are involved. Every action is logged, live systems are backed up before edits, and a daily summary arrives in chat so a person always knows what happened.
It is the same pattern we use in-house for our 14-bot SEO team, ads agent and site ops agent on the owner's own businesses (in-house systems, not client results). For a new team we start small, usually a specialist, a reporter and the team lead, and add agents only once the first ones are trusted. See how we build agent teams and real in-house use cases.
What should you prepare before talking to us?
You don't need a technical brief. A useful first conversation needs only a few things, and gathering them is valuable even if you never work with us:
- The process in plain words: what triggers it, who does what, and where it usually goes wrong.
- Rough volumes: how many messages, orders, tickets or invoices per week. A one-week tally is enough.
- Your tools: WhatsApp app or API, CRM, sheets, accounting or helpdesk software, including versions or plans where you know them.
- Three real examples: typical messages or cases, plus one awkward one, with personal details removed.
- What must never happen: the mistakes that would hurt most. These become the first lines of your stop rule.
- An owner: the person who will review the workflow once it runs.
Send these on WhatsApp or by email, or run the pilot scorecard first.
How do you know whether the pilot worked?
We agree the success measure before building, and it should be something your team can count without a dashboard project. Typical measures include the share of enquiries logged the same day, time to first response, the number of follow-ups that actually happened, hours spent on the task per week, or the number of escalations that were correctly flagged. We count the baseline before launch (even by hand for a week) and again after a few weeks of running.
We also review the qualitative side with the people involved. Do they trust the drafts? Are handovers arriving with enough context? Are customers asking for a person more or less often? A pilot can hit its number and still need tuning, or miss it and teach you something important. Either way, the decision to keep, tune, expand or stop is yours, made on evidence. We don't publish generic success figures, because results depend on your process, volumes and team.
How does a pilot with us work?
Every engagement follows the same four steps. Consult: you walk us through the process, the tools and where it breaks. Pick the pilot: we choose one workflow and write down its inputs, outputs, tools, stop rule and success measure. Ship: we build it into the tools you already use and test it on real cases. Review: your team checks what the agent did and what it escalated, then decides whether to tune it, expand it or stop. Read the full pilot process.