AI development: having AI applications built the right way

What does independent guidance for AI development cost, and who does the building?

Dave Willemstein doesn't build software, but guides AI development from the client's side: specification, build-partner selection, testing deliverables and governance. For choosing the build partner there's the Vendor Review over 3-6 weeks for €9,800 - €15,400, and during the build there's guidance from the client's side, for example as a Fractional AI Lead for €2,600 - €10,400 per month, for 2, 4 or 8 days per month. The build itself sits with the client's own team or with a development partner the client contracts directly.

AI development: the right service for each phase of the project

Phase of the project	What the client gets	Timeline	Price	Suitable service
At the start, when there's no clear picture yet of where the organisation stands	A logical first step, including a closing presentation to the leadership team or board	2-3 weeks	€7,000	AI Readiness Assessment
When choosing the build partner	Scorecards, a test on your own data and a three-year TCO comparison	3-6 weeks	€9,800 - €15,400	Vendor Review
During the build	Guidance from the client's side: testing deliverables, safeguarding the business case and governance	2, 4 or 8 days per month	€2,600 - €10,400 per month, for 2, 4 or 8 days per month	Fractional AI Lead

Is there a margin on developers from the network for AI development?

No. For AI development, the client contracts the build partner directly, with no intermediary layer, margin or commission. A party from the network goes through the same selection as any other party.

What determines the price of the Vendor Review for AI development?

The Vendor Review costs €9,800 - €15,400, depending on the number of vendors being compared.

What works best for AI development: build, buy or partner?

The hybrid form often works best for AI development: buying in building blocks, having the layer with your own knowledge built, and placing the implementation with a partner. Before that comes the question of whether building is needed at all: only build what sets the organisation apart in the market, and buy what's a commodity.

Source: rates and typical timelines from Dave Willemstein. Updated: October 2026.

What is AI development?

AI development is the design, building, testing and deployment of software in which AI plays a core role. Think of an application that reads documents and fills in fields, an assistant that searches internal knowledge, or an agent that carries out tasks in your systems. Three things set it apart from ordinary software development: the output isn’t always predictable, quality depends heavily on your data, and the system needs to keep being measured after it goes live.

That’s why good AI development doesn’t start with the developer, but with the question of which problem you’re solving and how you’ll measure whether it’s working. Only once that’s clear does a developer know what to build.

I’m not the builder in that picture. I work as an independent consultant on your side of the table: helping sharpen the brief, choosing the right build partner, and testing what gets delivered. The building itself is done by your own team or a development partner you contract directly. I don’t earn anything from the build, so I have no interest in a bigger scope.

Want to know which profile you actually need first? A separate article on what an AI developer actually does, day to day, sets out the different roles, from strategist to machine learning engineer.

What you can have developed

Most AI development questions fall into three categories. They overlap, but each calls for a different specification and often a different kind of build partner.

AI applications

An AI application is a standalone tool, often an AI web application, that supports a defined task: summarising case files, preparing quotes, classifying customer queries. Anyone having AI applications built usually buys in the underlying language model and has the layer around it built: the user interface, the rules, the connection to your data, and the evaluation. Which types exist and when custom development pays off is covered on the page about choosing, integrating or building AI software.

AI agents

An AI agent goes a step further: it carries out actions in your systems itself, such as raising a ticket or drafting and queuing an email. That calls for tighter agreements on permissions, logging and human oversight. AI agent examples sets out a dozen applications across customer service, sales, HR and finance.

Integrations with existing software

Often the most valuable form of AI development isn’t a new application, but AI inside the software you already use: your CRM, your document management system, your ERP. The AI component is then small; the integration, the permissions model and the data flows are the real work. An overview of applications, from document processing to internal assistants, is on custom AI solutions.

Development process

An AI project runs through roughly five phases. Below, for each phase, is who does the work and where I’m involved from your side.

1. 1. Brief and specificationYou and me: problem, success measures, constraints and a first build-buy-partner choice
2. 2. Build-partner selectionYou decide, I guide: criteria upfront, a level playing field, contract and exit terms
3. 3. Prototype on your own dataBuild partner builds, your team tests, I check against the agreed measures
4. 4. Into productionBuild partner and IT: integrations, security, logging and support; I watch scope and risk
5. 5. Support and measurementYour organisation owns it; periodic evaluation, adjust or stop

What you get from me in concrete terms depends on the phase:

• At the start: a specification in plain language, covering the problem, the success measures, the data needed, and the constraints around privacy and the AI Act. If you don’t yet have a clear picture of where you stand, the AI Readiness Assessment is a logical first step.
• When choosing the build partner: a Vendor Review with scorecards, a test on your own data, and a three-year TCO comparison. The investment is €9,800 - €15,400, over 3-6 weeks.
• During the build: guidance from your side, for example as a Fractional AI Lead (€2,600 / €5,200 / €10,400 per month), testing deliverables and safeguarding the business case and the governance around it.

What happens afterwards, from pilot to daily use, falls under AI implementation. That’s less about code and more about working processes, adoption and ownership.

An example of how that can look in practice: in the case on KYC document analysis for a financial services firm, the emphasis wasn’t on the model but on the working process around it and on guiding the vendor and implementation team.

Build versus buy

Before you have anything developed, the question of whether you should build it at all comes first. Only build what sets you apart in the market, buy what’s a commodity, and choose a partner if you want to steer the outcome without owning the technology yourself. The hybrid form often works best: buying in building blocks, having the layer with your own knowledge built, and placing the implementation with a partner. A full four-question decision framework for this choice is covered elsewhere on this site, in an insights piece on building, buying or partnering.

Working with developers from my network

Because I work alone and don’t build, I sometimes suggest parties I know for delivery: developers, data engineers or specialists in a particular platform. How that works is kept deliberately simple:

• You contract directly. There’s no layer through me. Agreements on scope, price, ownership and liability are made directly with the build partner.
• No margin, no commission. I earn nothing on their rate and hold no partner status with vendors.
• Known relationships are disclosed upfront. If I’ve advised a party before or worked with them, you hear that before they’re put forward.
• A suggestion isn’t a preference. A party from my network goes through the same selection as anyone else.

More on how I select partners, and what you can and can’t expect from them, is on the partner network page.

Want to talk through a development project that’s coming up, or a build partner who’s already sent a quote? Book a thirty-minute strategy call. At the end you’ll know what I’d do, even if the answer is that you don’t need me for it.