Topic guide

Agentic AI: Build vs Buy Decision Framework

When custom-built agents on the Agent SDK beat the commercial platform, and what each side actually costs.

The decision in five steps

Workload shape

Standard CX, sales, ITSM, legal? Buy a vertical agent. Bespoke domain or novel workflow? Build.

Team capacity

Do you have 2+ senior engineers free for 6 months? If not, buy. If yes, build is realistic.

Time to value

Need it live in 90 days? Buy. Have 6+ months? Build is on the table.

Compliance

Are tenant-isolated runtimes and audit trails a hard requirement? Some platforms ship that out of the box; building takes 3+ months.

Spend horizon

If platform cost over 3 years exceeds engineering cost of build, build wins. Otherwise buy.

Agent SDK (free)
Claude Agent SDK and OpenAI Agents SDK are free. The bill is what they consume.
LLM tokens
Anthropic Claude Sonnet $3 / $15 per 1M, OpenAI GPT-5.5 $5 / $30, Google Gemini 3.5 Flash $1.50 / $9.
Compute and hosting
Sandboxed code execution, browser-use, vector store. Roughly $1,000 to $5,000 per month for a small production deployment.
Engineering time
2 senior engineers at 50% allocation for the first 6 months, then 1 engineer at 25% for ongoing.
Eval and observability
LangSmith, Braintrust or Arize. $200 to $2,000 per month depending on traces.
Integration build
Connectors to your CRM, ticketing, knowledge base. Often the biggest hidden line.