Should you hire a developer to build your AI systems?
Hiring a developer looks like the serious choice and is usually the expensive way to learn what you should have built. Here is the arithmetic, using published developer wages and the telephony and model prices that arrive either way.
Lumina9 min read
Hire a developer when the software itself is what you are selling and it has to be different from anything you could buy. Rent a white-label stack when what you are selling is the outcome and the software is how you deliver it. Most one-to-ten-person AI agencies are in the second group and spend their first year behaving like they are in the first, which is how a promising agency ends up with a half-finished codebase and three clients.
That is the short answer. The longer answer is arithmetic, and the arithmetic is less flattering to hiring than most people expect, because the build quote is the smallest number in the whole conversation.
What does a developer actually cost?
Start with the public number. The US Bureau of Labor Statistics puts the median annual wage for software developers at $135,980 in May 2025, with the bottom 10 percent under $82,460 and the top 10 percent over $214,670. The same profile projects 10 percent employment growth from 2025 to 2035 and roughly 106,100 openings a year. Nothing in that says the price of this labor is about to fall.
Most small agencies do not hire full time. They contract, which feels cheaper because it is a project number instead of a salary number. It is priced off the same labor market, plus a margin for a contractor who carries their own downtime, and it buys you a fixed scope rather than a person who will still be there in March.
Here is the part that gets skipped. The quote covers the build. It does not cover the discovery calls where you work out what you actually want, the two rounds of revisions after you see it, the fork you need when client two books differently than client one, or the Saturday in August when the agent stops answering and the developer is at a wedding. We wrote about who maintains an AI system after it goes live because that bill is the one nobody quotes.
What hiring a developer does not remove from your bill
A developer builds the thing that spends the money. They do not buy the minutes. Every one of these lines still arrives whether you wrote the code or rented it.
Telephony. Twilio's US voice page, current as of August 2026, lists $0.0140 per minute to place a local call, $0.0085 per minute to receive one, and $1.15 a month per local number. Toll free receives at $0.0220.
Models. OpenAI's API pricing page lists GPT-Realtime-2.1 at $32 per million audio input tokens and $64 per million audio output tokens, with the mini version at $10 and $20. Streaming transcription through GPT-Live-Transcribe is $0.017 a minute.
Voice. ElevenLabs' pricing page puts the Pro plan at $99 a month for roughly 600 minutes of text to speech, with extra minutes around $0.17, and advertises low-latency output as low as 5 cents a minute on the Business tier.
Those are the raw inputs to one voice agent. Assembling them yourself is a real option and sometimes the right one, which is why we ran the full comparison in build your own AI stack, or use a white-label one. The point here is narrower. Hiring somebody to wire those services together changes who does the wiring. It does not change the invoices.
The number that actually decides it is client two
A developer building for your first client builds a single tenant. One business, one set of hours, one service list, one calendar, one voice.
Client two wants different hours. A different price band. A different way of triaging an emergency versus a maintenance call. If the first build was not designed for a second tenant, client two is a fork of the codebase, and client ten is ten forks, ten deploys, and ten places a bug can hide.
Multi-tenancy is the expensive engineering. Not the agent prompt. The prompt is the part that looks hard in a demo and is actually the cheapest thing in the system. The boring machinery underneath, per-agency subdomains, branded portals, isolated configuration, usage accounting, 100 plus vertical packs with real services and price bands written per trade, is what costs six figures to build and what nobody scopes on the first call.
Agencies usually discover this at client three. It is a bad time to discover it.
What you actually own in each case
Hire a developer and you own the repository. You also own the pager. Both of those are real, and people who say "at least I own it" are usually thinking about the first one.
Rent a platform and you own the client relationship, the brand on every surface, the retail price, and the contract. You rent the delivery layer. If that sounds like a weaker position, ask which of the two your client is actually paying for. They are paying for answered calls and booked jobs. They have never once asked to see the repo.
When hiring a developer is genuinely the right call
There are four situations where I would tell you to go hire somebody.
The workflow is yours and nobody sells it. You need an integration into a niche field-service system, a state licensing database, or a proprietary quoting engine, and no white-label product covers it. Pay for that one piece.
You have a repeated unmet ask. More than two paying clients have requested the same thing your platform will not do. That is a product signal, not a preference.
The arithmetic flips at your volume. Take your live client count, multiply by the wholesale monthly you pay, annualize it, and compare against a developer's fully loaded cost using the BLS figure above as a floor. If your annual wholesale spend is comfortably clear of that, and the system is stable enough to hand over, building becomes defensible.
The software is the product. If you intend to sell software rather than sell outcomes, you need to own it eventually. Just be honest about which business you are in, because the two require different skills and different investors.
Where renting is the wrong answer
The honest limits, since a comparison post without them is an ad.
The Lumina Sales platform is $49 a month and comes with 100 live voice minutes, 500 chat messages, 50 website audits, one published demo at a time, and a five minute cap on any single call. That is roughly forty demo calls a month. It is sized for selling, not for running a client's phone line. Client builds are metered separately.
The sandbox disclosure on a published demo is always visible. Your logo, colors, subdomain, and booking link are what the prospect sees, and Lumina appears nowhere, but the page does state that the AI is a demo and that bookings inside it are simulated. If your pitch depends on the prospect not knowing, this is the wrong product.
Client builds are priced on a 30-minute call rather than a public rate card, which some people hate. The published guides for what you pay us start at $550 setup and $125 a month for a website with a chatbot, and run to $1,150 to $1,400 setup with $315 to $460 a month plus $0.35 a minute on voice for the top tier. You set retail. The gap is yours. But you cannot quote your own price from our site before you talk to us, and if you need a fixed public wholesale number to sell against, that is a real friction.
And the obvious one. You are renting. Terms can change, and a vendor is a dependency in a way an employee is not. A developer you already paid cannot raise your monthly. Weigh that honestly. Our answer is that nothing is deleted if you cancel, published demos simply unpublish and everything stays as a draft, but that is a mitigation, not a refutation.
One timing note, since it is on the site right now. The $49 monthly rises on October 1.
A decision rule you can run in ten minutes
Write down how many clients you have live today. Live, not pipeline.
Zero to three: rent. You do not yet know what you are building, and a developer will happily build the wrong thing on your budget.
Four to fifteen: rent, and keep a written list of every request the platform cannot satisfy. That list is your future scope document and it costs nothing to maintain.
Sixteen and up with the same unmet request appearing repeatedly: get a quote for that one capability. Not for a rebuild.
The expensive mistake is hiring a developer to build the platform you have not yet proven you can sell. The cheap version of that experiment is renting one for a month and trying to close with it.
FAQ
Can I start on a white-label platform and build my own later?
Yes, and it is the normal path. What transfers is the client relationship, the pricing, the brand, and the intake data you collected. The code does not transfer, which is the point. By then you will know exactly what to ask a developer to build because you will have a list of things clients asked for.
Will my client know I did not build it?
Not from the demo. The published page carries your logo, your colors, your subdomain, and your booking link, and there is no Lumina branding on it. The one line always present is the sandbox disclosure stating that the AI is a demo and bookings in it are simulated. That line is deliberate. A demo that pretends to be a live system is a worse demo, because the first thing a curious prospect does is try to book something and then wonder why nothing arrived.
Is hiring a contractor cheaper than hiring an employee for this?
For a defined one-off piece, usually. For anything that needs to still work in six months, the comparison is not really contractor versus employee. It is one-off cost versus ongoing cost, and voice systems are firmly in the ongoing column because models, telephony, and the client's own business all keep changing.
What does a voice minute cost me once a client is live?
$0.35 a minute wholesale on builds that take calls, agreed on the 30-minute pricing call alongside the setup fee and monthly. The 100 minutes in the $49 plan are for your demos, not for client traffic.
Where to start
Do the arithmetic before you do anything else. Live client count, times wholesale monthly, annualized, against $135,980 as your floor for a developer. If renting wins at your current volume, and it almost certainly does under ten clients, then the only question left is whether the demo closes.
So test that. Talk to the receptionist on the live demo, ask it the question your next prospect will ask, and see whether it answers like the trade it is supposed to be. If it does, start an agency workspace and put it in front of someone this week. If it does not, you have saved yourself a developer.
