Engagement tiers
Discovery sprint
- Two-week working session
- Map of 3-5 highest-leverage AI opportunities
- Build vs buy recommendation per opportunity
- Prioritized roadmap with cost and timeline
- Working prototype on the top opportunity
- Senior team facilitation throughout
Production AI agent
- 60 to 90-day build
- Production-grade RAG or agentic system
- HubSpot or Salesforce native integration
- Eval suite and quality gates
- Observability stack (LangSmith, Helicone, or custom)
- Compliance review (SOC 2, GDPR alignment)
- 60 days of post-launch support
Multi-agent system
- 90 to 180-day build
- 3+ specialized agents with orchestration
- Shared knowledge layer (vector store + permissions)
- Tool-use across HubSpot, Salesforce, custom APIs
- Custom evals per agent + system-level evals
- Production observability + cost monitoring
- Senior architect named on engagement
- 90 days of post-launch support
- Senior AI engineers, no offshore hand-offs
- Fixed-fee against scope, no timesheet billing
- Weekly working session and demo
- Eval suite handed over with the system
- Cost monitoring and budget alerts in place
- Compliance review against your industry
- Documentation handed over at launch
- 30-minute strategy call before any commitment
Common questions
Why is AI work fixed-fee when most agencies do T&M?
Because AI engagements that bill hourly are how customers end up with $200K invoices for a chatbot. We've built enough production AI to scope reliably. The price you see is the price you pay.
What about LLM API costs?
Separate from our fee. You pay OpenAI / Anthropic / your cloud provider directly. We help you size the budget and add cost monitoring before launch.
Do you offer managed AI operations?
Yes. Managed AI ops starts at $7,500/month: monitoring, eval running, prompt updates, model upgrades, cost optimization. See /ai/managed-operations.
Can you work with our existing models / vendors?
Yes. We're model-agnostic. Most engagements use OpenAI or Anthropic; some use Bedrock or Vertex; some are mixed. We pick what fits the use case, not what's trendy.
What about hallucinations and bad outputs?
Eval suite is mandatory on every production engagement. Quality gates block bad outputs from reaching users. We don't ship without them.
Is this just prompt engineering?
No. Our average engagement is RAG + agentic patterns + integration + observability. Prompt engineering is one slice. The rest is software that has to work in production.
