Custom AI development starts from €3,000 excl. VAT at Codito Ergo Sum, quoted per project after a 1-to-2-week scoping phase. On the French market in 2026, the alternatives are the consultant billed at a daily rate (€800 to €1,500 per day) and subscriptions to managed agents — each model with its own logic and its own blind spots.
Try it yourself: search for the price of custom AI development. You'll find "contact us" pages, calculators that ask for your email before showing anything, and articles that answer "it depends". Everyone has a reason to publish nothing — and the executive trying to budget an AI project ends up comparing quotes that vary tenfold, with no framework to understand why.
This article lays out that framework. Not to sell a dream: so that you walk into any meeting — including with us — knowing how to read an AI development quote. The market's three pricing models and their real ranges, what objectively drives the price, our own pricing grid, the method, and the timelines we can honestly commit to.
The market's 3 pricing models
1. The consultant on a daily rate: €800 to €1,500 per day
The historical model. In France, in 2026, a senior AI consultant or developer bills between €800 and €1,500 per day depending on specialization and location — the average sits around €1,000–1,200 excl. VAT. A project of a few weeks adds up fast: 15 days of actual work means €15,000 to €20,000.
The advantage: flexibility — you buy time, you steer. The blind spot: the risk sits with you. If the project takes twice as long as planned, your budget absorbs it. And a daily rate pays for time spent, not results delivered — two things that don't always coincide. We broke down that math (and its hidden costs: onboarding, post-mission follow-up, knowledge leakage) in our comparison of the monthly AI Partnership vs the one-off consultant.
2. The agency fixed fee: quoted per project, varying tenfold
The dominant model for defined projects: one scope, one price, one set of deliverables. It's ours. On the French market, observed fixed fees for AI projects range from a few thousand euros for a focused agent to tens of thousands — sometimes far more at the large IT services firms — for complete business platforms. The gap doesn't only reflect project size: it also reflects the agency's cost structure (sales team, project management, offices) and how it prices in risk.
The advantage: budget predictability — overruns are the agency's problem, not yours. The point of caution: a fixed fee only makes sense if the scope has genuinely been defined. A fixed fee signed on a vague scope means either a cascade of change orders, or an agency cutting corners on quality to protect its price. Hence the importance of scoping — we'll come back to it below.
3. The subscription: managed agents
The emerging model: you don't pay for the development, you pay for the usage — a few hundred to a few thousand euros per month for an agent that is hosted, maintained and continuously improved. It makes sense for standardized needs (a support chatbot, automated monitoring) and it smooths out cash flow.
The trade-off: you own nothing. Stop paying and the agent disappears — and with it, sometimes, the data and business logic you've invested in it. For a core business process, the question of ownership and reversibility deserves to be asked before signing, not after.
What actually drives the price
Two requests that fit into the same sentence — "an AI agent that handles our inbound emails" — can legitimately vary fivefold. Here are the four variables that explain the gap, in the order they weigh:
- The complexity of the reasoning required. Sorting emails into three categories is a solved problem. Understanding a complete client file to draft a legally binding response is another matter entirely. The more nuanced the expected decision, the more orchestration, evaluation and guardrail work is required.
- The number and nature of integrations. A standalone agent is quick to deliver. An agent that reads your CRM, writes to your ERP and posts notifications to your messaging tool requires connectors, permissions, tests — every connected system adds work, especially if its APIs are dated or nonexistent.
- The state of your data. AI doesn't invent your knowledge: it leverages it. If your documents are structured and accessible, we move fast. If we first have to run OCR on ten years of scanned PDFs and reconcile three contradictory databases, that work has a price — and it conditions everything else.
- The level of interface. An agent that answers inside a tool you already use costs less than a full web app with authentication, roles, dashboards and payments. The application layer can account for half a budget.
Add to that the cross-cutting requirements — security, data sovereignty, GDPR compliance — which don't double a budget but are never free either.
And after delivery: the recurring costs
One point too many quotes stay silent on: an AI agent in production has running costs. Model consumption (billed per usage by the providers), hosting, and evolutive maintenance — models change several times a year, and an agent that is never touched degrades. For a focused agent with reasonable usage, we're typically talking a few tens to a few hundred euros per month; for a high-traffic platform, more. It's not a trap, it's a budget line: demand that it appear in the quote, with an estimate of the expected consumption. In ours, it always does — a development price without an estimated operating cost is not a complete price.
Our pricing: transparent, from €3,000 excl. VAT
At Codito Ergo Sum, custom AI development starts from €3,000 excl. VAT. Every project is individually quoted — not out of commercial coyness, but because the four variables above make any catalog price dishonest. The logic of the tiers, on the other hand, is public:
- The focused agent — one process, one agent, few integrations. This is the entry point, from €3,000 excl. VAT. Typical example: a monitoring or qualification agent that works on a defined source and delivers in a defined format.
- The integrated system — one or more agents connected to your tools (CRM, ERP, drive, messaging), with a dedicated knowledge base. The quote depends essentially on the number of integrations and the state of your data.
- The business platform — a complete web app with an AI layer: authentication, roles, interfaces, sometimes payments. This is the top of the range, and the scoping phase is what fixes its exact scope.
Three real examples from our portfolio to make these tiers concrete:
- Jarvis (BTP / engineering) — a multi-agent system for landscaping and VRD (roads and utilities) design studies: specialist agents (Landscape, VRD, Hydraulics, CCTP), each backed by a dedicated knowledge base, with plan and diagram reading (OCR + text), text or voice input, and exports to standard formats. Typically an advanced "integrated system".
- Alter Ego (communications) — an AI video avatar trained in a studio session, with voice cloning faithful down to the accent, and an automation that generates daily scripts embodied by the avatar. A focused agent on the surface, whose value lies in the quality of the fine-tuning.
- RailMove Connect (rail) — a platform matching freelancers with companies in the sector: prototype validated first, then development of the core features (profiles, assignments, multi-criteria filters, ratings, administration). The "business platform" tier, delivered in stages to de-risk the investment.
The method: scope before you price, deliver in iterations
Our process comes in four phases, and the first one explains all the others:
- Scoping (1 to 2 weeks). We map the target process, the state of your data, the required integrations, and we define the exact scope of the V1. That's what allows us to give a firm price — and what protects you from the vague fixed fee described above.
- Short sprints. Development moves forward in visible iterations: you see the agent working on your real cases along the way, not at final delivery. Adjustments happen while they're still cheap.
- Delivery into production. Not a POC gathering dust in a corner: a tool plugged into your systems, with your teams trained to use it. What gets built belongs to you.
- What comes next, at your pace. Maintenance, evolutions, scaling up — either on demand, or as part of our monthly AI Partnership for those who want a partner who stays.
Realistic timelines
Announcing timelines without knowing the scope would be as dishonest as announcing a firm price without scoping. But honest orders of magnitude do exist: 2 to 6 weeks for a focused agent, 4 to 12 weeks for a system integrated with your tools, and longer for a complete business platform — which we then break into milestones (prototype, V1, iterations) so that value arrives early. And the factor that most often stretches timelines isn't technical: it's the availability of data and access on the client side. Good scoping detects it before anything is signed.
Custom doesn't mean from scratch
One last misunderstanding to clear up, because it explains irrational quotes in both directions. "Custom" doesn't mean reinventing the wheel on every project: foundation models exist, the architectural building blocks (RAG, OCR, agent orchestration, connectors, authentication) are proven — we reuse them from one project to the next, and that's precisely what makes a €3,000 entry ticket possible rather than €30,000.
Custom is the assembly: your processes, your data, your tools, your business rules. It's the difference between a suit cut from existing fabric — nobody weaves their own — and a suit off the rack. The first fits you. The second fits everyone, and therefore no one.
Conclusion: demand a framework
The fog around AI development pricing isn't inevitable — it's often a commercial choice. Our position: an executive who understands why a quote costs what it costs is a better client, and a better long-term partner. Daily rate, fixed fee or subscription; complexity, integrations, data, interface: you now have the framework.
A process in mind, a budget to define?
Book 30 minutes with Emile — free, no commitment — for an honest first scoping of your custom AI development project.