Question 1

What's the typical timeline?

Accepted Answer

A discovery sprint takes one to two weeks. A working prototype follows in two to three weeks. Production builds typically run six to twelve weeks depending on integrations. We always ship a working prototype before committing to a full build — so you have evidence before signing the bigger check.

Question 2

Do you build custom agents or just integrate ChatGPT?

Accepted Answer

Both, depending on what actually solves the problem. Most engagements start with the question 'is an agent the right shape here at all?' If the workflow is deterministic, we build automation, not an agent. When an agent is the right call, we own the prompts, retrieval, tools, evals, and deployment — not just a chatbot wrapper.

Question 3

How do you measure success?

Accepted Answer

We define success criteria during discovery in finance-grade language: hours back, errors avoided, cycle time cut, conversion gained. Then we instrument the system to actually measure those metrics — not vanity model metrics that don't translate to outcomes.

Question 4

Will I have to learn anything new to maintain the system?

Accepted Answer

We default to mainstream stacks (TypeScript, React, Next.js, Firebase, etc.) so any reasonable engineering team can take over. We write runbooks for the operational team and onboarding docs for engineers. If something is intentionally niche (e.g., Power Automate, LangGraph), we make the trade-off visible during discovery.

Question 5

How much does an AI agent cost?

Accepted Answer

A discovery sprint starts around €4–6k. A prototype usually lands between €8–15k. Production builds depend on integration scope — most projects fall between €25–80k all-in. Retainer plans for ongoing operations start at €2k/month. We always quote firm before work begins.

Question 6

Why is discovery paid?

Accepted Answer

Because it's real work. We map systems, review data, talk to stakeholders, and produce an estimate grounded in reality. Free discovery would just bake in a margin elsewhere — we'd rather charge for the diligence and undercommit on the build.

Question 7

Do you do equity / revenue-share deals?

Accepted Answer

Rarely, and only in narrow cases. We're an engineering studio, not co-founders. If you need a long partner relationship with skin in the game, a retainer with clear scope works better than equity for both sides.

Question 8

What stack do you use?

Accepted Answer

TypeScript everywhere. React / Next.js for web. React Native / Expo for mobile. Tailwind + shadcn/ui for design system. Firebase, Supabase, or AWS for backend. Anthropic Claude and OpenAI GPT for LLMs. LangGraph for orchestration. Twilio for voice. Microsoft Power Platform when low-code accelerates delivery.

Question 9

Can agents work with Microsoft 365 and Power Platform?

Accepted Answer

Absolutely — it's one of our strongest stacks. We've shipped Power Apps and Power Automate solutions that integrate with custom AI services, including Dataverse-backed apps, custom connectors, and SharePoint-grounded chatbots.

Question 10

Which LLM do you default to?

Accepted Answer

We default to Anthropic Claude for reasoning-heavy work and OpenAI's GPT family for tool-calling and realtime voice — but we benchmark per project against open models (Llama, Mistral) when cost or data residency demands it. Vendor lock-in is explicitly avoided.

Question 11

Do you write tests and evals?

Accepted Answer

Yes, both. Standard unit + integration tests for the deterministic code, and eval suites for anything model-driven — golden test sets, regression checks, and drift monitoring tied to CI. Skipping evals is the fastest way to ship an agent that quietly degrades.

Question 12

How do you handle data privacy?

Accepted Answer

We default to deployments where your data never trains a foundation model — using API endpoints with zero-retention policies and, where regulation requires it, self-hosted or in-region open models. We sign NDAs and DPAs, document data flows, and keep secrets out of code.

Question 13

Do you sign NDAs and DPAs?

Accepted Answer

Yes — as standard. We use either your paper or our standard templates. For regulated industries (healthcare, finance), we walk through the data flow before signing so the agreement reflects the actual architecture.

Question 14

Where is the code and data hosted?

Accepted Answer

By default in your cloud — Firebase, Vercel, AWS, Azure, GCP — under your accounts. The code lives in your git from day one. We avoid SaaS lock-in by design.

Question 15

Do you handle hosting and maintenance?

Accepted Answer

Yes. We deploy on Firebase, Vercel, Cloudflare, or your own cloud, with monitoring, eval runs, and a retainer for ongoing prompt tuning, model upgrades, and the inevitable feature creep. You can also take everything in-house — we hand over docs, runbooks, and CI/CD configured to your team's preferences.

Question 16

What does handover look like?

Accepted Answer

A short handover sprint at the end of the build: runbook walkthrough, eval suite training, CI/CD walkthrough, and a 30-day grace period where we're available on Slack for questions. After that, you can keep us on retainer or run it in-house — your call.

Question 17

What if the model we use gets deprecated?

Accepted Answer

Model upgrades are a normal cost of ownership. Our eval suites are designed to catch regressions when we swap models, and our prompt code is version-controlled so rollbacks are trivial. Most upgrades are a 1–2 day exercise on retainer, not a re-build.

Questions, answered.

Send a short note, we'll reply within a day.