Home › Paths › Forward Deployed Engineering
Learning path
Forward Deployed Engineering
You sit with a customer, find the real problem and ship a working system in their cloud.
Why this path
A forward-deployed engineer is part engineer, part consultant. The skill is not one tool; it is getting from a messy conversation to a deployed, evaluated, costed system the customer's team can own. It draws on all the other paths.
How to use this page: 1. learn the key concepts, which are the same everywhere; 2. pick your platform; 3. work through Crawl, Walk and Run.
Read alongside: The AI Universe · Agent Harness, drawn simply · Eval Harness, drawn simply · Certification prep on this site.
Step 1 · the same everywhere
Key concepts
Learn these ideas first. They are the same on every cloud and with every tool; only the product names change. Each one is a short page: what it is, why it matters, an example and the common mistake.
Understand
Deliver
Scale
Step 2
Where do you work?
Pick your platform and you get one page with that platform's products in every step.
Beginner · 6 weeks
Crawl: You can build one thing end to end, with help.
Run discovery and turn a problem into a small, deterministic-first scope.
Key ideas at this level, the same on every platform:
Open this level on: AWS · Azure · Google Cloud · Databricks
Learn: the concept, and the product on each platform
| Learn | AWS | Azure | Google Cloud | Databricks |
|---|---|---|---|---|
| Discoverythe problem in their words; who feels the pain; what success looks like | Python | Python | Python | Python |
| Data mapwhere their data lives and who owns it | Glue Data Catalog + Lake Formation Amazon S3 | Microsoft Purview ADLS Gen2 / OneLake | Dataplex Cloud Storage | Unity Catalog Unity Catalog volumes on cloud storage |
| Deterministic-first scopewhat code can do; where a model is needed | Amazon Bedrock | Azure OpenAI in Microsoft Foundry | Gemini on Vertex AI | Foundation Model APIs |
| Writinga one-page plan and a weekly update | GitHub | GitHub or Azure Repos | GitHub | GitHub + Git folders |
Build on GitHub
Forward-deployed discovery kit
Templates for week one at a customer: the problem in their words, a data map, what is deterministic, what needs a model.
Your first agent, three tools
A support assistant that can look up an order, check a policy and draft a reply, where code checks every tool call first.
You are ready to move on when
- You can write the problem in one paragraph the customer agrees with.
- Your scope says what the model will NOT do.
- You have a plan with a date on it.
Intermediate · 3 months
Walk: You can build it so it keeps working, without help.
Prototype fast on real data and deploy into the customer's cloud.
Key ideas at this level, the same on every platform:
Open this level on: AWS · Azure · Google Cloud · Databricks
Learn: the concept, and the product on each platform
| Learn | AWS | Azure | Google Cloud | Databricks |
|---|---|---|---|---|
| Prototype in daysa working demo on their data, not slides | Lambda / ECS Fargate Amazon Bedrock | Container Apps / Functions Azure OpenAI in Microsoft Foundry | Cloud Run Gemini on Vertex AI | Databricks Apps Foundation Model APIs |
| Their cloud, their rulesidentity, networking, approvals | IAM + Secrets Manager Terraform or AWS CDK | Entra ID + Key Vault Bicep or Terraform | IAM + Secret Manager Terraform | Unity Catalog grants + secrets Declarative Automation Bundles (Asset Bundles) |
| Prove itevaluation and cost the customer can read | Bedrock evaluations Cost Explorer + CUR | Foundry evaluations Cost Management | Vertex AI Gen AI evaluation Billing export to BigQuery | MLflow evaluation Billing system tables |
| Data to answerpipelines into the use case | AWS DMS / AppFlow dbt on Redshift or Athena | Data Factory copy / Fabric mirroring dbt on Fabric | Datastream Dataform or dbt on BigQuery | Lakeflow Connect / Auto Loader Lakeflow Declarative Pipelines or dbt |
Build on GitHub
One-week prototype on real data
Take a public dataset that looks like a customer's, build a working demo in five days, and write the hand-over note.
Claims assistant on Azure
A motor-insurance claims agent running for real: tools, checks, approvals, durable waits, evaluation and cost.
Eval Harness Framework
Rebuild each conversation from the agent's own traces and check it against rules written in plain YAML.
You are ready to move on when
- Your prototype runs in their account, not yours.
- You can show quality and cost on one page.
- Their engineers can run it without you.
Advanced · 6 months +
Run: You can design it, prove it and lead others through it.
Lead engagements: reusable assets, hand-over and many customers.
Key ideas at this level, the same on every platform:
Open this level on: AWS · Azure · Google Cloud · Databricks
Learn: the concept, and the product on each platform
| Learn | AWS | Azure | Google Cloud | Databricks |
|---|---|---|---|---|
| Reusable assetsone framework, many customers, bindings per cloud | Terraform or AWS CDK Bedrock AgentCore | Bicep or Terraform Foundry Agent Service / Durable Functions | Terraform Vertex AI Agent Engine | Declarative Automation Bundles (Asset Bundles) Mosaic AI Agent Framework |
| Hand-overrunbooks, training, ownership | CloudWatch + X-Ray | Azure Monitor + App Insights | Cloud Logging + Trace | System tables + MLflow tracing |
| Assuranceevidence that it is safe to run | Bedrock evaluations Bedrock Guardrails | Foundry evaluations Azure AI Content Safety | Vertex AI Gen AI evaluation Model Armor | MLflow evaluation AI Gateway guardrails |
| Commercial sensescope, estimate, cost to run | Cost Explorer + CUR | Cost Management | Billing export to BigQuery | Billing system tables |
Build on GitHub
Port the agent to your cloud
Take the reference agent and bind it to AWS or Google Cloud: same code, new adapters, same tests passing.
Reference agent
A customer-support agent with no framework: one module per harness layer, the architecture enforced by tests.
AskSherlock
An open-source ask-your-data assistant over a cost warehouse, with every answer traceable to a metric.
You are ready to move on when
- You have handed a system to a team that now runs it.
- Your second customer took half the time of the first.
- You can estimate the monthly running cost before you build.