FDE Sprint: Become a Forward-Deployed Engineer in 6 Weeks

Who it is for

For software engineers. Two groups get the most from the Sprint.

Change your role

Engineers who want an FDE role

You have 3+ years of experience in backend or full-stack work. You ship production code now. You want an FDE, applied-AI or AI solutions role.

Become AI-native in your role

Senior engineers and engineering leaders

You have 10+ years of experience. You want to build agentic AI systems yourself and to lead AI delivery at your company. You take back to your work: a software factory, an eval set, architecture decision records and a security review pack.

Do you have less than 3 years of experience? You can join if you can build and debug software in a backend or full-stack language. Read more ↓

Not the right fit if you are not a software engineer, you want a recorded course, or you cannot keep Tuesday and Thursday evenings free.

Why this program

Most FDE courses teach the technology stack. Employers hire for the customer work.

A forward-deployed engineer works inside the team of a customer. The FDE finds the real problem, builds the AI system, puts it into production and gets people to use it. We studied what FDE job postings ask for, and we built the Sprint around it.

8x

growth in US FDE job postings in a year, from 643 to 5,330

70%+

of FDE postings ask for customer discovery, which most courses do not teach

6 of 10

FDE postings ask for evals, the technical skill that employers test most

Sources: Indeed data via Paraform (Apr 2025 → Apr 2026); Perspective AI analysis of 1,000 FDE postings, 2026; theCRUX research on FDE job postings, 2026.

What you will be able to do

Four outcomes. Each one gives you proof of work.

You do each part in a simulated client engagement. It is a case study that runs through the full program. It recreates the unglamorous, difficult problems of enterprise AI.

Find the real problem

You will be able to lead a discovery interview and write a scope, so that you build the correct system.

You leave with scored interview transcripts and an approved scope of work.

Build an enterprise-class system

You will be able to build RAG, an agent and an MCP server on the Claude stack, and deploy them behind SSO.

You leave with a deployed system, not a prototype.

Prove that it works

You will be able to write evals that a client trusts, so that each change is safe.

You leave with a 50-case eval set, a red-team set and a regression eval.

Get people to use it

You will be able to pass a security review, give an executive readout and plan adoption.

You leave with a security review pack, a readout and an adoption plan.

Your instructors

Taught by people who have been the customer.

Shameek Chakravarty

Co-founder · Lead instructor

20+ years across Amazon, Yahoo, Sun Microsystems and Juniper, then founder of Farmizen. Previously built two agentic AI businesses at a Silicon Valley AI venture studio. CS, BITS Pilani · MBA, ISB.

LinkedIn →

Bharanidharan Viswanathan

CEO & co-founder · Lead instructor

Three-time founder: 91mobiles, SaaSworthy and Intoglo. Went deep on AI maths and fundamentals in 2021; now runs an AI venture studio. CEG, Anna University · MBA, ISB.

LinkedIn →

How the six weeks run

Intense by design. Four stages, three gates, one Demo Day.

50 hours of live sessions and about 50 hours of self-paced work. All sessions are live on Zoom.

5–6 Dec · 14 h live

Foundations weekend

You write a spec, build with Claude Code, deploy an app and monitor it. You do the full cycle in two days.

8–22 Dec · Tue and Thu

Discover, scope, start to build

You do discovery interviews and get your scope approved at Gate 1. Then you build RAG and your first agent.

5–21 Jan · Tue and Thu

Harden and deploy

You write evals and guardrails, deploy behind SSO and pass Gate 2. Then you do the security review and the executive readout.

25–30 Jan

Capstone week and Demo Day

Your pod completes its graded capstone project. On Sat 30 January, your pod demos it to a panel.

Also included: 6 hours of office hours, 6 hours of Claude Certification Exam Prep Workshops and a bonus 1-1 mock interview with an industry expert. There are no sessions from 24 December to 3 January. See the full schedule ↓

The details

Do you want more detail? Open the part that you want to read.

The full schedule, session by sessionA Foundations weekend, ten live working sessions, three gates and Demo Day

A gate is a checkpoint. Your pod must pass it before you continue. Thus you find problems before the holiday break, not on Demo Day.

When	Session	You ship
Sat–Sun 5–6 Dec	Foundations weekend, 14 hours live: the FDE role, then spec to production with Claude Code	A working app with a written spec and tests, deployed and monitored
Tue 8 Dec	Claude API for production	A classifier with structured outputs; accuracy before and after on 50 samples
Thu 10 Dec	Tool use and MCP	An MCP server; your first discovery interviews
Tue 15 Dec	Scoping and decision models Gate 1	An approved scope of work and escalation memo; your ticket classifier rebuilt on a System One decision model (such as TypeSafe Jev) and compared with Claude
Thu 17 Dec	RAG and context engineering	Cited answers over a 200-page document corpus; a context budget for your agent
Tue 22 Dec	Agents, MCP and A2A	A triage agent on the Claude Agent SDK
24 Dec – 3 Jan	Holiday break	Bonus: Claude Certification Exam Prep Workshop for the Developer exam (Tue 29 Dec, 7:30–9:30pm) and self-paced modules
Tue 5 Jan	Evals 1: evals the client trusts	A 50-case eval set from client-style data, with a baseline, error analysis and a calibrated LLM judge
Thu 7 Jan	Deploying, local models and fine-tuning	Deployed behind SSO with tracing; an open model running locally with Ollama, LoRA fine-tuned and compared with Claude on your eval set
Tue 12 Jan	Evals 2: guardrails, red-teaming and regressions Gate 2	A red-team set, a regression eval that runs on every change, a planted failure in an unfamiliar repo caught and fixed with a post-mortem, and a demo that clears the eval bar
Tue 19 Jan	Security, readouts and adoption	A security review pack (questionnaire, threat model, data-handling answers), an executive readout and an adoption plan
Thu 21 Jan	Architecting AI systems: lifecycle and enablement	Architecture decision records for your build, a lifecycle plan (versioning, model upgrades, drift, rollback) and an enablement kit for the client's engineers
25–29 Jan	Capstone week	No sessions. Your pod completes its capstone project for Demo Day
Sat 30 Jan	Demo Day Gate 3	Each pod demos its capstone project as running software in front of a panel, then individual questions
Thu 4 Feb · Tue 9 Feb	Bonus Claude Certification Exam Prep Workshops for the Architect exams	Architect: Foundations, then Architect: Professional, 7:30–9:30pm

A normal week, and the hoursTwo evenings each week, office hours and self-paced work

The full program is online. You join live Zoom sessions with the instructors and do self-paced work between sessions. Nothing is in person. The program starts with a Foundations weekend. Sessions are on Tuesday and Thursday evenings, with office hours on Mondays. The last week before Demo Day is a capstone week with no sessions. The program ends with Demo Day on Saturday 30 January. The Sprint is six weeks of program time. With the holiday break, it runs from 5 December to 30 January. Plan for 10 to 12 hours each week: two 2-hour evening sessions, a 1-hour office hour and 5 to 7 hours of self-paced work.

14 h

Foundations weekend

Two days, 5–6 December

20 h

Live working sessions

10 sessions of 2 hours, on Tuesday and Thursday evenings

4 h

Demo Day

Saturday 30 January: each pod demos its capstone project

6 h

Office hours

Mondays 8–9pm, live with an instructor

6 h

Claude Certification Exam Prep Workshops

Three 2-hour workshops: one in the holiday break, two after Demo Day

50 h

Self-paced work

About 50 hours on pre-reads, assignments and your capstone project

50 hours of live sessions and about 50 hours of self-paced work: about 100 hours in total, all online. You also get a bonus 45-minute 1-1 mock interview with a practising industry expert, and 12 months of online AI tutors and Lab practice exams.

Tue · 7:30–9:30pm

Teach and build

Opens with a decomposition drill on a vague brief.

Thu · 7:30–9:30pm

Teach and build

Opens with a customer role-play: explain it to a CFO, push back on scope.

Mon · 8–9pm

Bonus office hours

Bring your build and your blockers.

Your pod

Four engineers, mixed seniority

We set the pods. Each member owns a workstream: retrieval, agent and MCP, evals or deployment.

Every 2-hour session

15405510

15 minWarm-up drill or role-play 40 minTeach and live demo 55 minBuild in your pod 10 minAssignment with a measurable target

Every assignment is scored on Works, Measured and Explained. Grading is individual.

Why this program is differentWhat FDE job postings ask for, and where you do it in the Sprint

3 things that make the FDE Sprint different.

1 · We read the job postings

We teach what FDE employers ask for

Our research covers thousands of FDE job postings, from companies such as Anthropic, OpenAI, Palantir, Scale AI, Sarvam, Glean, Decagon, Harvey, Sierra and Composio. They ask for evals, customer discovery, scoping, enterprise integration, adoption and field feedback to the product team. Most courses teach the technology stack and stop there. We teach the technology stack and these skills.

2 · Built on the Claude stack

You prepare for Claude certification while you build

You build on Anthropic's Claude stack: the Claude API, the Agent SDK, MCP and Claude Code. The sessions map to all four Claude certifications, so you prepare for the exams as you do the work.

Anthropic currently opens the exams only to people at Claude Partner Network organizations. If your employer is a partner, you can sit them.

3 · Intense by design

A six-week change of direction

The Sprint is for engineers who want to reinvent themselves: for an FDE role, or as an AI-native engineer in the role they have. We do not like long programs that lose energy. The Sprint is intense by design: a 14-hour Foundations weekend, two evenings each week, three gates, a capstone week and a public Demo Day. It is hard work for six weeks. Then you have the result.

What the job postings ask for	Where you do it in the Sprint
Evals. In 6 out of 10 FDE job postings. It is the technical skill that employers test most.	Evals 1 and Evals 2: a 50-case eval set, a calibrated LLM judge, a red-team set and a regression eval.
Customer discovery and scoping. “Own discovery, technical scoping, system design, build, and production rollout.” (OpenAI)	Scored discovery interviews from the Foundations weekend. Gate 1 is an approved scope of work.
Push back on the wrong ask. “Turn vague asks into specs, and push back when the ask is wrong.” (Composio)	A decomposition drill each Tuesday and a customer role-play each Thursday.
Field feedback to the product team. In 7 out of 10 FDE job postings, and in none of the FDE courses that we compared.	On Demo Day you hand in a one-page reusable deployment pattern.
Adoption and handover. “Run training, write crisp runbooks, and hand off durable playbooks.” (Harvey)	An executive readout, an adoption plan and an enablement kit for the client's engineers.

Sources: theCRUX research on FDE job postings and FDE course pages, 2026. It uses published analyses of thousands of FDE postings (Indeed data via Paraform, 5,330 postings; Perspective AI, 1,000 postings) and our own reading of postings from the companies named. Quotes are from the job postings of the companies named. These companies have no connection to theCRUX.

The terms that you will be able to explain20 terms from FDE job postings, each with the session where you build it

We found these terms in FDE job postings at AI labs and startups. Each term maps to a session and to a thing that you ship. In an interview, you explain the term from your own work, not from a definition.

Spec-driven development

Write the spec and tests first, then let Claude Code build against them.

Foundations weekend · spec + tests

Claude Code

An AI coding agent you use to build, test and debug, including in code you did not write.

Foundations weekend, S8 · planted-failure drill

Fine-tuning (LoRA, QLoRA)

Adapting an open model to a task, and knowing when prompting or RAG beats it.

S7 · fine-tune vs Claude on your evals

Local models

Running open-weights models on your own machine or the client's servers with Ollama, for when data cannot leave the building.

S7 · Ollama deployment

Decision models

"System One" models such as TypeSafe Jev return typed decisions instead of text, for fast, cheap classification and routing.

S3 · classifier vs Claude

Loop engineering

Build self-improving loops: the system checks its own output against evals and uses the failures to improve the next run.

Foundations weekend, S8 · software factory, regression eval

Context engineering

Deciding what goes into the model's context, what gets compacted and what gets remembered.

S4 · context budget

RAG

Fetch the right documents so the model answers from the client's data, with citations.

S4 · cited answers

Evals

Tests that score answers on every change. The skill employers ask for most.

S6 Evals 1 · S8 Evals 2

Agents

A model that calls tools and takes steps on its own, and when not to use one.

S5 · triage agent

MCP

A standard plug that connects a model to a client's systems.

S2 · MCP server

Observability

Tracing every model call and tool step so you can debug and see cost.

S7 · tracing dashboard

Guardrails

Blocking prompt injection, PII leaks and off-topic answers.

S8 · red-team set

Security review

Answer the security questions of the client: a questionnaire, a threat model and data-handling answers.

S9 · security review pack

Customer discovery

Finding what the client actually needs before you build.

Foundations weekend, S2 · scored transcripts

Scoping

Success criteria, what is in, what is out, and saying no.

S3 · scope of work

Enterprise integration

SSO, secrets, data residency, and working inside the client's network.

S7 · deployed behind auth

Hallucinations

Why models make things up, and how citations and evals catch it.

S4, S6, S8

Token economics

Know what each model call costs in tokens, and control the cost with the choice of model and the size of the context.

S3, S4, S7 · decision model, context budget, cost tracing

Adoption

Getting real users to use what you built: runbooks, training, ROI.

S9 · readout

Claude certification exam detailsThe four exams, the workshops and who can sit the exams

Anthropic runs four proctored Claude certifications. The Sprint's sessions map to all four. For each exam you get 12 months of online AI tutors and Lab practice exams. You also get three bonus 2-hour Claude Certification Exam Prep Workshops. The Developer workshop is in the holiday break. The two Architect workshops are after Demo Day, when your engagement is complete.

Foundations · covered in the Tue and Thu sessions, plus a bonus Claude Certification Exam Prep Workshop

CCAO-F · Associate: Foundations

Guiding customers to the right use case

For consultants, sellers and delivery leads.

Sessions: discovery, scoping and the readout (Foundations weekend, S2, S3, S9)

CCDV-F · Developer: Foundations

Building with the API, Claude Code and MCP

Every learner should be ready to sit it by Demo Day.

Sessions: S1 API, S2 MCP, S4 RAG, S5 Agent SDK, S8 Claude Code · Workshop 1 (Tue 29 Dec): exam walkthrough and practice questions

Architect · covered in the sessions, with two bonus Claude Certification Exam Prep Workshops after Demo Day

CCAR-F · Architect: Foundations

Designing Claude solutions end to end

Your engagement is the case study: you design it, build it and defend it.

Sessions: S3 design and scope, S5 agents, S6 and S8 evals, S7 deployment, S10 architecture decisions · Workshop 2 (Thu 4 Feb): exam-style design case reviews

CCAR-P · Architect: Professional

Enterprise architecture, evals and governance

The hardest of the four; senior engineers get the most from it.

Sessions: S6 and S8 evals and red-teaming, S7 local models and fine-tuning, S9 security and governance, S10 lifecycle and enablement · Workshop 3 (Tue 9 Feb): a timed practice exam and debrief

FDE Sprint participants get access to practice material for all four exams and our AI tutor Sherpa.

Straight answer on access: Anthropic currently opens these exams only to people at Claude Partner Network organizations, and exam fees are paid by you. We prepare you to pass; we cannot guarantee a result or book you a seat. If your employer is a partner, you can sit them.

How we grade youDiscovery scores, three gates and a mock interview

FDE interviews test decomposition. You get a vague problem and 45 minutes, and you must make a plan. You practise this every Tuesday, and we score it in all six weeks.

Grading is individual. We grade each learner on their own discovery interviews, their own assignments and their own answers on Demo Day.

A client that hides the real problem

An operations head, an IT lead and a frontline user who want different things
Hidden facts that only come out when you ask the right question
A curveball in week 3: a dependency you planned on gets banned
Every interview transcript scored on what you uncovered

An instructor plays the ops head at Gate 2 and at the executive readout.

A bonus 1-1 mock interview

45 minutes with a practising industry expert: a vague brief, scoping under pushback, no code. You get a written scorecard within 48 hours.

The mock interview is optional. Senior engineers who do not need it can skip it.

The six bonusesIncluded in the price

The Sprint fee includes each bonus. The office hours and the workshops are part of the 50 hours of live sessions.

Bonus 1 · 45 minutes

1-1 mock interview

A 45-minute 1-1 mock interview with a practising industry expert. You get a vague brief and you scope it while the expert pushes back. You get a written scorecard in 48 hours.

Booked between 18 and 29 January

Bonus 2 · 6 hours

Monday office hours

One hour each Monday, live with an instructor. Bring your build and your blockers.

Mondays 8–9pm IST · 7, 14, 21 Dec · 4, 11, 18 Jan

Bonus 3 · 6 hours

Claude Certification Exam Prep Workshops

Three 2-hour workshops for the Developer and the two Architect exams. You also get 12 months of online AI tutors and Lab practice exams for all four exams.

Tue 29 Dec · Thu 4 Feb · Tue 9 Feb

Bonus 4 · Self-paced

Holiday modules

Short self-paced modules for the holiday break. There are no required sessions in the break.

24 Dec – 3 Jan

Bonus 5 · Self-paced

Engineering refresher

AI/ML basics, with exercises. Use it in the pre-work week if you need it.

Pre-work week, 28 Nov – 4 Dec

Bonus 6 · Plugin

FDE job search plugin

A Claude Code plugin for your FDE job search. It covers the full workflow: it finds FDE job postings, ranks them against your profile, prepares your résumé and your application for each role, and tracks each application.

Works in Claude Code

More on who it is for, and what you needExperience levels and the subscriptions that you pay for

Two groups of engineers get the most from the Sprint. They want different results, and the same six weeks give both.

Change your role

Engineers who want an FDE role

You have 3+ years of experience in backend or full-stack, in any language. You ship production code now. You want an FDE, applied-AI or AI solutions role.

Become AI-native in your role

Senior engineers and engineering leaders

You have 10+ years of experience, and you see your work change because of AI. You want to build agentic AI systems yourself, such as an agent that triages real tickets, and to lead AI delivery at your company. Your years with customers, stakeholders and production systems are half of the FDE job. The Sprint adds the other half. You take back to your work: a software factory, an eval set, architecture decision records, a lifecycle plan and a security review pack.

Also welcome

Strong engineers with 0–2 years

You must be able to build and debug software in a backend or full-stack language. You join a pod and own a workstream. The Sprint prepares you for deployment and solutions engineer roles on an FDE track.

Everyone builds in the Sprint, and uses coding agents such as Claude Code extensively. Pods have mixed seniority, so engineers at different levels work together.

Not the right fit if you are not a software engineer yet, want a recorded course, or cannot keep Tuesday and Thursday evenings. Start with Claude for Life instead.

What you will need: a laptop that can run Claude Code, a Claude Pro subscription (or higher), an Anthropic API key from the Claude Platform with some credit loaded, and a free Google Colab account for the fine-tuning lab. These are in your name and paid by you; they are not included in the Sprint fee.

Questions

Answers before you join

Is the client engagement real?

No. The client engagement is simulated. It is a case study that runs through the full program. It teaches you the difference between a prototype and an enterprise-class system that works and gives business value. AI agents play the client team.

What happens after I register?

You pay online and get a confirmation by email in a few minutes. Pre-work starts on 28 November: you set up Claude Code and your API key. The Foundations weekend is on 5–6 December. Registration closes on 1 December, or when 24 seats fill.

Do you promise a job?

No. This is the founding cohort, so there are no past learners yet. The Sprint gives you proof of work and practice for the FDE interview.

How much time does it take?

About 100 hours: 50 hours of live sessions and about 50 hours of self-paced work on pre-reads, assignments and the capstone project. Plan for 10 to 12 hours each week: two 2-hour evening sessions, a 1-hour office hour and 5 to 7 hours of self-paced work. The Sprint is intense by design.

Is grading individual?

Yes. We grade each learner on their own discovery interviews, their own assignments and their own answers on Demo Day. Each assignment is scored on Works, Measured and Explained.

Do I need AI experience?

No. You must be a software engineer who works in a backend or full-stack language. A pre-work week teaches the Claude API basics.

What do I need to pay for, apart from the fee?

A Claude Pro subscription and an Anthropic API key with some credit. You also need a laptop that can run Claude Code.

Does the Sprint prepare me for the Claude certifications?

Yes, for all four. Anthropic currently opens the exams only to people at Claude Partner Network organizations. You pay the exam fees.

Can my company pay?

Yes. We give a GST invoice. Teams of four or more engineers can build the capstone around their own real problem.

FDE Sprint · cohort 1

Six weeks. One simulated client engagement. Proof of work that you can show.

The Foundations weekend is on 5–6 December 2026. On 30 January 2027, your pod demos its capstone project to a panel on Demo Day. Register now to hold your seat. Registration closes on 1 December, or when 24 seats fill.

Register now See the six weeks

Questions? Email hello@thecrux.ai