In short
- Most hours in a small agency go on work that repeats: updates, small changes, reports and chasing answers.
- Hand over jobs that recur and can be checked. Keep the clients, the decisions, the design judgement and anything you cannot undo.
- Write your house rules down. An agent, like a new hire, only knows what is written.
- Approve every plan before work starts and ask for proof after it, so quality does not depend on who did the job.
- Sell retainers and outcomes, not hours, and hire for the work that needs a person.
Where the hours go in a small agency
Ask an agency owner where the week went and the answer is rarely the work they sold. If you want to scale a web agency without hiring first, start by counting where the hours actually go. For most small agencies it is five things:
- Maintenance. Backups, checks, and fixing what stopped working. See website maintenance.
- Updates. Plugin, theme, platform and PHP versions, plus security patches.
- Small changes. A new price, a swapped banner, a paragraph moved, each one arriving as an email.
- Reporting. The monthly report that looks the same every month and still takes an afternoon.
- Chasing answers. Waiting on a login, a decision or an image, and keeping track of who owes what.
None of this is the work clients remember you for. All of it still has to be done, and it grows with every retainer you sign. That is why capacity feels fixed: the repeating work grows with the client list, and the work you enjoy gets squeezed into what is left.
The hire-first trap
The usual answer is to hire a developer. It is a fair answer, but it is slow and it costs more than the salary. Recruiting takes time. The new developer then needs time to get as quick as you on your clients' sites, and meanwhile you are the one showing them the stacks, the logins and the way you like things done. Add equipment, holiday cover and your own management time.
Then comes the trap. You hire to clear a backlog of repeating jobs, and the hire spends most of the week on those same jobs. You have added a person and a fixed monthly cost, and the work that stops you growing is exactly where it was. The next retainer needs another hire.
Hiring fixes a headcount problem. Most small agencies have a repeating-work problem first.
What to hand over first, and what to keep
Hand over jobs that pass two tests. They happen again and again, and you can prove the result. A plugin update followed by a check that the pages still load passes both. A decision about what a client's brand should feel like passes neither.
| Job | Hand it over? | Why |
|---|---|---|
| Plugin, theme and platform updates, with a check afterwards | Yes | It repeats, and a before-and-after check proves it. |
| Small changes that arrive by email | Yes, with a plan you approve | The request is clear and the result is visible. |
| Monthly reports | Yes | The same shape every month. |
| Follow-up emails to clients | Draft only, you send | The relationship stays with you. |
| Agreeing scope and price | Keep | It is a commercial judgement. |
| Design direction and brand decisions | Keep | It is the judgement the client is paying you for. |
| Anything you cannot undo, such as deleting data or switching a live shop | Keep the go-ahead | An agent can prepare it. A person decides. |
Here is a simple test for a first job. If you gave it to a new starter with a checklist, could you tell whether they got it right? If yes, it is a good candidate. If you would have to watch them do it, keep it for now. Our guides to AI for web agencies and AI website maintenance go deeper on which jobs suit it.
From our records: moving 27 hosts from PHP 8.0 to 8.2
Northgate client needed 27 hosts moved to a newer PHP. It is the kind of job that repeats, and it has a clear before and after.
- Before the switch, the agent saved the status and the page size of every host.
- After it, 25 of 27 hosts came out byte-identical, which is close to proof that nothing changed.
- The same baseline showed 1 site returning 200 with a 310-byte page while its siblings returned about 150,000. It had been broken for about 6 months, and no status check could see it.
- One homepage moved from 168,244 to 168,249 bytes and nobody could say why, because only the size had been saved.
The rule we wrote down: save the page, not just its size.
Example based on a real job, details changed.
Write your house rules down
A new hire learns your house rules by getting them wrong. An agent can be told them in advance, but only if they exist in writing. How do you test on staging first? Which clients need a ticket before any change? What do you never do on a live shop on a Friday? If the answers live in your head, neither a new hire nor an agent can use them.
So write them down, one short rule at a time, each with the mistake that caused it. That is how the Webgentic Orchestrator works. Every night it reviews the day's jobs and turns real mistakes into written rules. The well-evidenced ones reach every agent, and bigger changes wait for a person to sign them off. We have 861 of these rules so far, as of 29 October 2026, each written up with what to avoid next time. You can start with a plain document today, and it will help your next hire too.
From our records: 14 answers nobody wrote down
On a call with Larch client, we walked a numbered list of open questions one at a time and settled 14 of them out loud. Nobody wrote the answers back. Eight days later the list still showed all 14 as open, and the client had twice told a colleague that all the questions were answered.
Nothing had been lost. The file's history showed that nobody had overwritten anything, so the gap was capture, not data. The rule we wrote: write a call's answers back the same day, with the time they were said, and keep the hedge ("tentatively", "I would say"). An answer that lives only in a recording looks open to everyone who checks.
Example based on a real job, details changed.
One agent per client
A person carries every client in their head, which is how clients get mixed up. The Orchestrator gives each client its own agent and each project its own agent, with its own memory and its own sandbox. A project agent can only read and change its own project's files, and that limit is enforced in code, not left to good behaviour.
For you, that means the agent for a client already knows the stack, the contacts and your rules for that account. Big jobs are split between helper workers inside the same project, and the project agent checks their work before it reports back.
Plans before work, proof after
In a small agency, quality often depends on who did the job. Your best developer gets it right, and a tired Friday afternoon does not. Two habits take the person out of it: a plan approved before work, and proof after it.
In the Orchestrator, every change to a client site starts as a short numbered plan with the risk and the way back. A person approves it, changes it or rejects it, and nothing touches a client site before then. Every job ends with a written validation of what changed, what was tested and what is still open. Visible changes come with screenshots, checked point by point.
From our records: five days of errors after one upgrade command
On a Magento shop for Acme client, an upgrade command wiped the compiled front-end files. Nobody rebuilt them, so the shop returned a server error for 5 days. When the storefront came back, the admin screens were broken, because only one language had been rebuilt and the admin uses another.
The rule we wrote: rebuild both languages after every upgrade, and do not call a deploy done until the storefront, the admin, a product page and a category page all load.
Example based on a real job, details changed.
The Orchestrator is not the only product we make, so a note on the other one. Webgentic for sites covers a single site. The owner asks in plain English, sees the plan, and watches the change appear. There is no approval step there: you can stop a job, read the history in plain English and restore any earlier version in one click. In the Orchestrator, a person approves each plan before work starts.
Price outcomes, not hours
Once agents do part of the work, hours stop being a sensible thing to sell. A routine update that takes minutes would make your best work your cheapest. Sell the outcome instead: the sites kept patched and backed up, the report delivered, a change made within an agreed time.
Price it as a retainer per site or per client. Work out what each kind of job costs you to deliver, then price above it with room to spare. The Orchestrator shows what every job cost to run, which keeps the sums honest. Keep hourly or day rates for new project work, where the scope is not yet known. Your price should not fall because the work got quicker. The client is paying for the result, and for you being accountable for it.
When you should hire anyway
Agents do not replace everyone. Hire when:
- The work needs a specialist skill you do not have, such as complex custom development.
- The role is client-facing, such as account management, where the client wants a person.
- Your pipeline needs a person: more enquiries than you can answer, or work that depends on relationships.
- You are the only reviewer. Someone has to approve plans and check proof, and if that is only you, a second senior person is worth hiring.
The aim is not to avoid hiring for ever. It is to spend each hire on work only a person can do.
How to scale a web agency without hiring: a 30-day start plan
- Days 1 to 3: pick 1 client and 1 recurring job. Choose one with a clear before and after, such as monthly updates on a single site.
- Days 4 to 10: write the rules. List how you do the job today, the logins and checks, and every mistake you have made at it. One short rule each.
- Days 11 to 20: run it with plans and proof. Have the agent write a plan, approve it, and read the proof after. At first, read every plan and every proof.
- Days 21 to 25: measure. Compare the hours, the mistakes found and the cost of each job with how you did it before.
- Days 26 to 30: add the next job or client. Same steps, starting from the rules you already wrote.
After a month you will know whether it works for your agency. The Orchestrator is installed by our developers in Bromley, who load your clients, access and house rules and train your team on real jobs, and it is priced on a call. Read how setup works, or see what the Orchestrator does.
