AI Agents & Automation
Custom AI agents
AI agents that work with your existing tools within the permissions you set: multi-step workflows, human approvals, guardrails, error handling and monitoring.
The problem
Tasks that need judgement
Some work does not follow fixed steps and still takes up time.
Off-the-shelf tools do not fit
Generic assistants cannot use your systems or follow your rules.
What is included
Agent design
One clearly defined job, with the tools and information it may use.
Tool access and permissions
Exactly which systems the agent may read or change.
Approvals and guardrails
Human approval for actions that matter; checks on what the agent does.
Logging and monitoring
Every step recorded, so you can see what happened and why.
An example, step by step
Input
The task
Each Monday, check stock levels for twenty items and prepare reorders for those below the minimum.
- Read stock levels and supplier terms with read-only access.
- Draft reorders; a request above the agreed limit is denied and flagged.
- A person approves the drafts; only then are the orders placed.
Output
A traced run
A log of each step: what was read, what was proposed, one action denied for exceeding the limit, and who approved the rest.
How it works
Define the job
We agree what the agent does, and what it must not do.
Build and test
The agent is built and tested on real cases before use.
Run with oversight
It runs with approvals and logging; we review its actions.
What you receive
A working agent
For one defined job, connected to the agreed tools.
Guardrails and logs
Permissions, approvals and a record of every action.
Handover notes
How it works, how to change it and what to watch.
What it works with
The tools the agent may use
Each system and action listed and agreed; nothing beyond it.
Your rules
Limits, approvals and what must never happen.
A test environment
Where the agent is tried before it touches live systems, when one exists.
How we check it works
Task completion
Checked on a set of realistic test cases, including awkward ones.
Policy adherence
Forbidden actions are refused in testing, every time.
Recovery and approval
Errors stop the run safely, and actions that need approval wait for it.
What to keep in mind
- An agent can make mistakes; actions with consequences keep a human approval.
- Agents are built for one job at a time; a broader scope means more testing first.
What shapes the scope and quote
- The number of tools and actions the agent needs.
- How strict the approvals and limits must be.
- Whether a test environment exists or has to be set up.
Results
Results from client projects
Published only with each client's approval.
Frequently asked questions
What is the difference between an agent and a workflow?
A workflow follows fixed steps. An agent decides the next step within limits you set. We use an agent only where that flexibility is needed.
Can an agent handle customer calls or support?
Voice and support agents are possible applications, built and tested for your case; they are not a ready-made product we sell.
Can the agent do things we did not allow?
It can only use the tools and actions it is given. Others are not available to it, and attempts to exceed limits are refused and logged.
Who is responsible for what the agent does?
You decide what it may do and approve what matters. We design it so every action is logged and can be traced back.
Learn more
Lead automation walkthrough
How a new website enquiry becomes a qualified CRM record and a reply drafted for approval.
How we work
How Wishfy works: discover, design, build, and measure and improve, for search visibility, automation and knowledge assistants.
Security & data
How Wishfy handles client data: a GDPR processing agreement, EU/EEA processing by default, no training on your data, and deletion at the end.
Other services in AI Agents & Automation
Lead capture & follow-up
New enquiries captured, qualified and added to your CRM, with follow-ups drafted for approval.
Inbox & CRM automation
Email sorted, replies suggested and CRM records kept up to date, with approval before sending.
Document & invoice processing
Fields read from invoices and other documents, checked, reviewed where uncertain and exported.
Reports & dashboards
Recurring reports collected, checked, summarised and delivered on a schedule.