A practical playbook for businesses to decide which agents to build vs. buy—hit outcomes faster, stay swap-ready, and avoid lock-in.

‍TL;DR

Build what sets you apart.

Buy what scales.

Use the scorecard below to choose, prove it with clear metrics over the next year, and keep your stack “swap-ready” with open protocols so you can change vendors without rewiring.

Core vs. context

Some work makes your business different in the eyes of customers. That’s core. The rest is context—table stakes you must do well but won’t win on.

“Companies bring in ‘core’ functions that differentiate them… and outsource the ‘context’ that’s table stakes to get right.” — Aaron Levie

In practice, the durable pattern is a capable front-of-house agent that steers conversations, with smaller task agents behind it. Don’t rebuild parity. Build core. Buy coverage.

The decision rule

If customers would notice—and it moves a top metric this year—lean to build.

If it’s about reach, speed, or standard operations—lean to buy.

This is a business decision, not the CX or Sales department call. Leadership, CX, product/engineering, ops, security, and finance should make it together.

The scorecard (print this and use it)

How it works: Score each item.

0 = no, 1 = maybe, 2 = strong yes.

Add them up. This gives you a bias toward build or buy. Revisit the score annually (most SaaS terms are yearly).

1. Differentiation. A great version of this would clearly set us apart.
2. Moves a top metric soon. It can lift NRR, cut churn, raise adoption, or improve CSAT within 12 months.
3. Proprietary logic or data. It encodes know-how or signals we won’t hand to a vendor.
4. Risk and brand. Failure or data leakage here would be hard to forgive (PII, compliance, safety).
5. Ownership capacity. We have people to own prompts, policies, evaluations, incidents, and model drift for the next 18–36 months.
6. Integration surface. It will touch many systems (CRM, product, billing, support, telemetry) and other agents.
7. Time to impact. We need results this quarter (this pushes toward buy).
8. Dedicated team. This agent will have a standalone team, not compete with other engineering work.
9. Reuse. The logic will be reused across segments, regions, or products.
10. Switching pain later. Even with open protocols, a vendor swap here would be painful.

Add it up:

• 15–20+ → Build (or partner-build and keep the IP).
• 8–14 → Buy now, measure hard, and re-score in one year.
• ≤7 → Buy products that can be configured deeply, customized, and personalized.

Guardrails either way (put these in the contract):

• You own inputs and outputs.
• No training on your data without explicit opt-in.
• Clear retention windows; sub-processor notice; export and audit logs on request.
• Identity and consent scoped by Account • Role • Purpose.
• Demand swap-ability: require open, documented protocols and interfaces (e.g., A2A for agent hand-offs, ACP for routing/policy, MCP for data/tools, plus human-approval hooks and a bridge for legacy apps) and require bulk data export, standardized auth, and the right to replace a component without touching upstream/downstream code.

The cost to own (the resource reality)

1. Owning an agent is not “build it once.” It is a service you run, update, and maintain over years.
2. You will keep changing prompts and policies as the business changes.
3. You will run evaluations and watch for model drift.
4. You will handle fallbacks, on-call, and post-mortems when things go wrong.
5. You will upgrade models and tools and pass new security reviews.
6. You will field a steady stream of “small changes,” because the agent touches many teams.
7. When you build, all of this competes with the rest of your roadmap.
8. When you buy, you get shared improvements and a clear service commitment, while you still own outcomes—not the plumbing.

“The risk that companies rebuild their systems of record is low… The real risk is better, AI-first versions beating legacy SaaS.” — Bucco Capital

Price the total cost to own and run over 12–24 months, not just the next sprint.

Business patterns (useful examples)

• Businesses often build when the agent holds their secret sauce.
• They build pricing and optimization logic that shapes margins.
• They build risk or triage logic that governs decisions.
• They build demand planning and allocation that balances service and cost.
• They build sourcing and savings heuristics that drive competitive wins.
• They build any agent where customers would feel the difference.
• Businesses often buy when the job is broad coverage and speed.
• They buy customer-facing agents for onboarding briefings, live Q&A, feedback loops, and renewal or expansion readiness.
• They buy staff-facing agents for quick briefs, support autopilot, and health summaries.

If you are operations-heavy, start by buying and configure with your data and policies.

Graduate to build only when customers would notice, you have a data advantage, and the scale or risk clearly justifies owning it.

You can also take a middle path: add small tools around a bought agent for your unique steps, and keep human approvals in the loop where needed.

‍

Cast Example

On the left and right are vendor-supplied agents that give you reach—customer presentations, Q&A, feedback, readiness, and internal briefs. In the center are the agents you build: the pricing, risk, planning, or savings logic that make you different. All agents interoperate over standard protocols (A2A/ACP/MCP with human approvals), so you stay swap-ready.

Build in the middle, buy at the edges—connect everything over open protocols.

A short note on standards

Whatever you choose, favor open protocols and interfaces—so agents can hand off work to each other, follow policy, reach data and tools safely, and escalate to people when needed. This keeps you able to swap parts without ripping out everything around them.

Prove it with simple metrics

• If you build: watch gross margin, loss ratio, Mean Time To Resolution (MTTR)‍ / first-time-fix, and forecast accuracy.

• If you buy: watch the percent of contacts reached, activation/adoption, OES/CES/CSAT, and renewal or expansion readiness.

Track before/after for one year. Let numbers—not opinions—settle the debate.

Close

Build your core. Buy your coverage.

Use the scorecard. Put the guardrails in your contracts. Choose standards that make change easy. That’s how you move fast—and stay free.

‍