What We Do
Every engagement follows the same three-phase model. It's designed for one outcome: a fully operational agentic team your engineering, finance, or ops team owns and runs — without calling us.
Diagnose
Every engagement starts here. We run a structured audit of your workflows, data, and team to identify where agentic AI creates the most leverage — whether in engineering, finance, or ops. Output: a prioritized implementation roadmap with effort estimates and expected ROI per initiative.
Duration: 2-4 weeks
Deliverable: Agentic Roadmap
Build
We design, build, and test your agentic systems alongside your engineers. Not black-box delivery — your team is embedded in every sprint. Stack is standardized (LangGraph + LangSmith + Postgres + AWS) and documented to the component level.
Duration: 8-16 weeks
Deliverable: Running systems, full codebase
Transfer
The engagement ends when your team is fully operational. We deliver runbooks, architecture decision records, LangSmith eval suites, and training sessions. You run it. We stay on call for 30 days.
Duration: 2-4 weeks
Deliverable: Runbooks, evals, training, 30-day hypercare
Our Stack
We're opinionated because we've made the tradeoffs so you don't have to.
| Layer | Technology | Why |
|---|---|---|
| Agent framework | LangGraph + LangChain | Production-proven, auditable, debuggable |
| Observability | LangSmith | Traces, evals, prompt management |
| Memory and retrieval | Postgres + pgvector | Open-source, self-hostable, no vendor lock |
| Tool protocol | MCP | Emerging industry standard |
| Cloud | AWS | Largest talent pool; broadest AI service catalog |
| IaC | Terraform | Most widely recognized; easy client handoff |
| Language | Python 3.12+ | Largest LLM/AI ecosystem; type-safe with mypy |
Frequently Asked Questions
What is an agentic AI system?
Software where AI models autonomously plan and execute multi-step tasks using tools — without a human directing each step. Examples: automated research agents, finance close automation, internal workflow automation, AI-powered customer operations.
How long does an engagement take?
12–24 weeks total: Diagnose (2–4 weeks), Build (8–16 weeks), Transfer (2–4 weeks). Exact timeline depends on workflow complexity and scope defined in the Diagnose phase.
Who owns the code and IP?
You do. All code, prompts, evals, architecture records, and runbooks live in your GitHub repo. Agentic Runbook retains no license or access after Transfer is complete.
What is co-build?
Your engineers work alongside ours in every sprint. They review every design decision and own the codebase from day one. No knowledge leaves with us when the engagement ends.