An Agent Does Things. That Is the Whole Difference, and the Whole Risk.
A chatbot answers a question, and the worst thing it can do is say something wrong. An agent takes an action — sends the email, updates the record, books the slot, issues the refund, tells a customer something they will hold you to. The interesting question was never what it can do. It is what it is allowed to do without asking anybody, and almost nobody writes that down before the build starts.
No credit card required · 30-minute call · No obligation to sign up
Six permissions somebody has to sign off
Each of these is a business decision that ends up encoded in software. Left undecided, it becomes whatever the build happened to do first.
- May it send, or only draft?The single biggest one
- May it change a record?Or only propose the change
- May it commit you to anything?A price, a date, a promise
- May it spend money?A refund, a credit, a booking
- What is the ceiling?Above which a person always looks
- What happens when it is unsure?Escalate, stop, or guess
The last row is the one that separates a system you can leave running from one you cannot. An agent that guesses when it is uncertain will be right most of the time, and the times it is wrong will be exactly the unusual cases — the ones that matter. Uncertainty has to route to a person, and that rule has to be written before anything runs on real work.
Start it as a drafter. Promote it when it has earned it.
The safest way to build an agent is to give it no authority at all to begin with. It reads, it decides, it prepares the action — and a person presses send. That looks like a compromise and it is actually the fastest route to a system you trust, because you spend two weeks watching what it would have done without any of those decisions being real. The mistakes it makes in that period are free, and they are the ones you would otherwise have found on a customer.
Then permissions get raised one at a time, with a ceiling. It may send its own replies to routine enquiries but anything above a value threshold still gets a person. It may update a record but not delete one. It may book but not cancel. Every one of those is a sentence somebody in your business has to agree to, and writing them down is most of what makes an agent safe rather than clever.
The second thing that matters is that everything it does is legible afterwards. Not a log file — a record a non-technical person can read that says what it received, what it decided, why, and what it did. When something goes wrong, and it will, the difference between an hour and a week is whether that record exists. The same record is what lets you widen its permissions with any confidence at all.
And the boring engineering point that gets skipped: an agent acting on your systems needs its own credentials, in your accounts, with the narrowest access that lets it work. Not a shared login, not a director’s account, and nothing with more reach than the task requires. The blast radius of a mistake is set at this stage, months before the mistake happens.
- ✓Start with zero authority — It drafts, a person sends. Two weeks of free mistakes.
- ✓Raise permissions one at a time — With a ceiling above which a person always looks.
- ✓A record a person can read — What it got, what it decided, why, what it did.
- ✓Its own narrow credentials — Not a shared login. This sets the blast radius.
- ✓Uncertainty routes to a human — Guessing is right most of the time and wrong on the cases that matter.
An agent that guesses when it is unsure will be wrong on exactly the unusual cases — which are the ones that matter.
What building an agent actually involves
Six things, and the first is a conversation rather than code.
A Written Permission Set
What it may do alone, what it may only propose, and the value ceiling above which a person always looks. Agreed by somebody in your business, in plain sentences, before anything is built.
Drafting Mode First
It prepares, a person presses send, for as long as it takes to trust it. The mistakes it makes in that window cost nothing and are the ones you would otherwise find on a customer.
An Escalation Rule
What it does when it is not confident: hand to a named person, stop, or ask. Never guess. Written into the build rather than left to whatever the model does under pressure.
A Readable Decision Record
What it received, what it concluded, why, and what it did — readable by somebody non-technical. It is the difference between an hour and a week when something goes wrong.
Its Own Narrow Credentials
In your accounts, with the least access that lets it work. Not a shared login and not a director’s account. This is where the blast radius of a future mistake is decided.
Reviewed After It Is Live
What it handled, what it escalated, and what it got wrong — read on a schedule rather than when somebody complains. Permissions widen from that, or narrow.
What clients say
Real clients, quoted in their own words — published with their permission.
“Ghalib’s clear guidance has helped improve our website’s SEO performance.”
“It was a pleasure working with Ghalib and his team.”
“We’re very thankful to Ghalib and his team for building a fully-functioning website for our business.”
More of them, in full, on our reviews page.
— Our Proprietary Methodology —
The Visibility Framework™, applied in AI Agents
The method doesn’t change by market. What it’s pointed at does.
Audit The Hours
Where time actually goes, task by task, scored on volume, repetition and the cost of getting it wrong. Ends in a ranked blueprint with estimated hours saved — yours to keep either way.
Design The Guardrails
Before any building: what the agent may touch, where a human must approve, what happens when it is unsure, and which data is never allowed near a third-party model.
Build & Evaluate
One workflow at a time, in your accounts, scored against real examples from your business before it touches live work. Shipped early so it meets reality while it is still cheap to change.
Run & Improve
Monitored for cost, failures and quality drift. Models change, your business changes, and an automation nobody tends becomes a liability rather than an asset.
Honest, No-Nonsense Commitment
If the audit concludes that a task is not worth automating, we will tell you and refund the difference rather than build it anyway. And if a workflow we built does not hit the outcome we agreed in the blueprint, we keep working on it at no extra cost until it does or we take it out.
AI Automation pricing for AI Agents, in GBP
Quoted in pounds after the permission set exists, because an agent priced before anybody has decided what it may do is priced for a different thing. Running costs — the model provider’s usage — are billed to your own account, not through us.
Single Agent
One job, one set of permissions.
- Written permission set, agreed first
- Drafting mode until you trust it
- Readable decision record from day one
Agent & Systems
The usual one. Acting across your tools.
- Everything in Single Agent
- Narrow credentials per system it touches
- Escalation routing to named people
- Optional: bundle in SEO or Website Development
Multi-Agent
Several agents, or one with real reach.
- Per-agent permissions and ceilings
- Scheduled review of what each handled
- Evaluation before every permission increase
- Optional: combine all 4 services for full-funnel growth
Every plan is scoped around your market — start with a free first look and we’ll recommend what fits, priced in GBP.
What you’re actually committing to
Most agencies keep this in a contract you only see after the sales call. We would rather you knew now, because it is the question everyone asks second — right after the price.
- The audit is credited, not sunkPay for the audit, and the full amount comes off the build if you proceed. If you don’t, the blueprint is still yours to hand to anyone else.
- A fixed build price after the auditQuoted once we know what we are building. If it takes longer than we estimated, that is our risk — the price only moves if you change the scope.
- You own everythingAccounts, API keys, workflows, prompts, evaluation sets, logs and documentation — all in your name from day one, and still yours if we never work together again.
- Running costs are yours and visibleAPI usage is billed by the provider directly to you. We never resell tokens or mark up usage, and you see the real number.
- The retainer is month to month30 days’ notice, no exit fee. Stop it and your automations keep running — you are simply maintaining them yourself.
- Human approval is the defaultAnything customer-facing or irreversible needs a person until the evaluation data justifies otherwise, and that decision is yours to make, not ours.
These are the terms as they appear in the agreement itself — nothing here is softened for the website. The full wording lives in our terms and conditions, and you get the agreement to read before anything is signed or invoiced.
AI Agents AI Automation questions, answered
Related AI Automation pages
Same service, different angle — by market, by service and by industry.
Ready to decide what it may do?
Free 30-minute call · No obligation · Response within 24 hours
Get in touch with Ghalib Ashrafi HQ
Ghalib Ashrafi takes on work for brands in six markets — the UK, USA, UAE, Saudi Arabia, Australia and Pakistan. Every enquiry gets a reply within 24 hours.