Forward Deployed Engineer
Not just an AI engineer — one who thinks from the outcome.
Forward Deployed Engineering is what OpenAI, Anthropic, and Palantir do with their most important customers: a senior engineer becomes part of your team — with real ownership of the result, not just the code. The difference: an FDE thinks in business impact and ROI, guides AI initiatives from the first idea to production, and ensures solutions actually get adopted.
from €200/h net
Forward Deployed Engineer — what does that actually mean?
A Forward Deployed Engineer (FDE) is an experienced software engineer who doesn't advise from the outside, but works directly inside your codebase, your team, and your context. The term comes from the US market — Palantir coined it, and OpenAI, Anthropic, Scale AI, and Lovable adopted it.
In the US, the model is standard at every AI-first company. In the DACH region, it's still emerging — which is exactly why it's an opportunity for the first companies to use it.
What that means for you: you get a model that's standard at the most successful AI companies in the world — at a time when your competitors are still working with classical consulting.
vs. traditional consultant
- writes code, not slides
- sits in the same chat (e.g., Teams) as the team
- knows the codebase, backlog, stakeholders
- delivers features, not recommendations
- thinks in ROI and business impact
- drives adoption, not just deployment
vs. body-leasing freelancer
- owns architecture decisions
- senior expertise, not just hands
- documents and trains
- point of contact for strategy
- asks: is this worth building?
- knowledge transfer as explicit goal
Forward Deployed Engineer vs. software team
An FDE doesn't replace every team — but for most mid-market AI initiatives, it delivers faster impact, less overhead, and clearer ownership. Click a row for details.
Time to impact
Forward Deployed Engineer
Productive in days, not months
A senior engineer with context is quickly in your repo, backlog, and team chat. No 3–6 month hiring cycle, no onboarding marathon.
Classic software team (3–5 people)
Months until the team is in place
Recruiting, onboarding, team dynamics, and first joint releases take time — especially for senior AI roles.
Cost & flexibility
Forward Deployed Engineer
Variable cost, clear scope
You pay for delivered senior capacity — from a sensible minimum commitment, scalable up or down anytime.
Classic software team (3–5 people)
High fixed costs from day one
3–5 full-time roles (backend, frontend, DevOps, PM, possibly ML) quickly mean €300–600k/year — even when less output is needed.
Coordination
Forward Deployed Engineer
One point of contact, low overhead
No daily meeting layers between roles. Decisions happen faster because architecture and implementation sit together.
Classic software team (3–5 people)
More alignment, more latency
Meetings, handoffs, and different priorities between roles slow decisions — especially in the first months.
Architecture & ownership
Forward Deployed Engineer
End-to-end ownership
I take responsibility for architecture, code quality, and product decisions — not just ticket execution.
Classic software team (3–5 people)
Distributed responsibility
Good for large products — but without clear AI leadership, everyone can build their lane and nobody owns the whole picture.
Business impact
Forward Deployed Engineer
Asks first: is this worth building?
Scope, ROI, and adoption get clarified before building. No feature without a business case.
Classic software team (3–5 people)
Often builds what was planned
Teams deliver reliably — but without a strategic filter, a lot can get built that nobody uses.
Knowledge transfer
Forward Deployed Engineer
Explicit learning goal
Pair programming, documentation, and explanatory sessions are part of the work — your team becomes more self-sufficient.
Classic software team (3–5 people)
Knowledge stays in silos
Each role knows their area — understanding the overall AI architecture has to be built actively.
Scale
Forward Deployed Engineer
Ideal for focused AI initiatives
One clear feature, one production push, one architecture sprint — fast and lean.
Classic software team (3–5 people)
Stronger with parallel streams
If you're developing several large product lines in parallel, a dedicated team can make more sense.
Rule of thumb: if you need a concrete AI feature live in weeks, FDE is usually the better choice. If you're scaling several large product lines in parallel, a team can make more sense — ideally with clear AI leadership.
What I ask first — before writing a single line of code
Is it worth it?
Before we start building, I clarify: what's the concrete business value? What changes — in processes, costs, customer experience? I don't build features that have no impact.
What actually needs to ship?
Not everything that sounds good needs to be in version 1. I help with scope-setting: what's MVP, what's nice-to-have, what's technical debt?
Who uses it afterward?
A feature nobody uses has no ROI. I think about adoption from day one: who are the users? What do they need? How does the knowledge get into your team?
"AI features don't fail because of the code. They fail because nobody needs them — or nobody knows how to use them."
What an FDE brings
Business & Product
Product-oriented thinking
What drives ROI? What's nice-to-have? I prioritize by business impact, not technical elegance.
Architecture & product ownership
I take responsibility for decisions — not just implementing tickets.
End-to-end ownership
From the first idea through architecture, implementation, and deployment to actual adoption by the team.
Knowledge transfer
Your team learns along the way — not just your system. After the engagement, you're more self-sufficient than before.
Tech Stack
LLM integration
OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, local LLMs — with a focus on production-readiness.
RAG systems
From embedding strategy to eval pipeline — with measurable quality KPIs.
Agentic workflows
LangGraph, MCP servers, tool calling, multi-agent — designed for real business processes.
AI-native features
Chatbots, co-pilots, classifiers — embedded in your existing web app, not bolted on.
DevOps for AI
Deployment, observability, LLM logging, cost tracking, and latency optimization.
Refactoring & modernization
Legacy codebase, migration to modern stacks, AI-assisted refactoring with Cursor, Claude Code, and Copilot.
"The most common mistake in AI projects: building the right thing wrong, or the wrong thing right. My job is to prevent both."
— Joshua Heller
Cancel anytime, no notice required — from a minimum commitment so the model makes sense for both sides.
A clear model
Hourly rate
- €200/h net — from a minimum of roughly 20 hours per month
- Why a minimum? FDE means building context — your codebase, your team, your goals. Under 20 hours per month, the model doesn't make sense for either party
- Need just a few hours for an assessment? AI Sparring (from €300/h) is the right path
- Cancel anytime, no notice required
- Ideal for: concrete features, architecture, ongoing AI development in your codebase
As a retainer
- Fixed monthly hour bank for predictable capacity
- Volume discount on the hourly rate, depending on scope
- Predictable, billed monthly
- Ideal for: ongoing codebase partnership over several months
When FDE makes sense — and when it doesn't
You have an AI feature that needs to go live in 4–12 weeks.
Your team can't do it alone in that time — I don't just build alongside them, I make sure it actually gets used.
You have a working prototype but no path to production.
From a Streamlit demo to a scalable system is a different mountain. I also check: is the prototype the right thing, or does the scope need rethinking?
You want to build AI expertise without 6 months of recruiting.
Instead of hiring a senior AI engineer, you bring me in for 3–6 months. Your team learns in every sprint — that's part of the deal, not a bonus.
You know AI can change things — but not exactly where or how.
I help identify the right initiatives and prioritize by real business impact.
When it doesn't fit:
- You primarily need strategy without implementation → Fractional CAIO is a better fit
- You don't have a direction yet → AI Workshop is the first step
- You're looking for pure development capacity without the strategic element → I'm happy to refer you
How we get started
Intro call
We check whether the FDE model fits your situation.
⏱ 30 min
Discovery
Code walkthrough, stakeholder calls, initial recommendations.
⏱ 1–2 weeks
Proposal & contract
Hour bank, scope, start date.
⏱ a few days
Ongoing engagement
Your tools, weekly calls, code in your repo.
⏱ as needed
"An FDE isn't the external consultant looking in from the outside. He's the senior engineer you don't currently have on your team — on loan."
— Joshua Heller
Frequently Asked Questions
What sets an FDE apart from a regular freelancer?
An FDE takes ownership — for architecture, code quality, and decisions. A regular freelancer works through tickets. I do both depending on context, but the default is: I think along, not just execute.
How many hours per week will you be available?
Standard is 10–20 hours, which works for most setups. During intensive phases, up to 30–40. I never go full-time on a single engagement — otherwise you lose the FDE advantage.
Do you only work on AI projects that are already clearly defined?
No — a large part of my work is exactly the opposite: figuring out which AI initiatives actually make sense. I help with scope-setting, ROI assessment, and whether the planned approach is the right one. That often saves more than any hour of implementation.
What happens if an AI initiative has no ROI in your assessment?
Then I say so. An honest conversation before the engagement beats six months of development for a feature nobody uses. My business model runs on referrals — short, successful projects beat long, unclear engagements.
How do you ensure our team can work independently after the engagement?
Knowledge transfer isn't a bonus — it's part of the work. Every sprint includes documentation, pair programming, and explanatory sessions. By the end, your team should understand the architecture, follow the decisions, and be able to continue developing — without me.
Where and how do you work?
Primarily remote, on-site regularly throughout the DACH region. I work with your tools — Microsoft Teams, GitHub, Cursor, Claude Code, Notion — and I'm open to your cloud (Azure, AWS, Supabase, Google Cloud). Everything digital, everything documented.
Are you alone, or do you bring a team?
In FDE engagements, I am the person. For larger engagements, I bring in TAISC engineers by agreement — but never without asking you first.
What if the engagement isn't working after 4 weeks?
You can stop at any time without notice — you only pay for hours delivered. There's a sensible minimum commitment for the model to work. My business model runs on referrals — an unhappy client costs me more than a canceled contract.
Do you have references?
Yes, available on request with a confidentiality agreement. Past FDE-style engagements: BCT Technology (Logify), i40 (SkillDuck), Bitwerft, Onventis.
Sound like what you need?
Let's talk for 30 minutes. If FDE doesn't fit, I'll tell you that too.
Prefer to write first? joshuaheller@theaisoftwarecompany.com