Build vs buy AI agents: how to decide in one meeting
Buy when the workflow is generic and identical at every company. Build when the workflow is your differentiator, touches proprietary data, or carries compliance weight. The one-line test: if the workflow shows up in your P&L by name, build it. Most companies should run roughly 80 percent bought tools and 20 percent built systems, with the built 20 percent producing most of the return.
When buying wins
Off-the-shelf AI wins on anything commodity. If a thousand other companies do the workflow the same way you do, someone has already productized it cheaper than you can build it:
- Meeting transcription and notes
- Email drafting and individual writing assistance
- Basic support deflection on common questions
- Generic research and summarization
Typical cost: $20 to $100 per user per month, live this week. Buying here is not lazy. It's correct. Paying custom prices for commodity capability is how budgets die.
When building wins
Build when the workflow meets any of these four conditions:
- It's the differentiator. If the workflow is why customers pick you, renting the same tool as your competitors caps your edge at zero.
- It touches proprietary data. Your pricing history, your compliance records, your operational patterns. Generic tools can't see them and shouldn't.
- It must act, not chat. Agents that update records, route work, enforce rules, and trigger downstream systems need integrations no SaaS vendor will build for you.
- It carries compliance weight. Audit trails, residency, and review gates have to be designed in. Details: AI in regulated industries.
The 3-year math
Per-seat pricing looks small until you multiply it. Run the comparison before any vendor meeting:
| Option | Year 1 | 3-year total | You own it? |
|---|---|---|---|
| SaaS tool, $50/user/mo, 100 users | $60,000 | $180,000 | No. Price and roadmap are theirs. |
| Custom agent, $90k build + 20%/yr maintenance | $90,000 + $18,000 | ~$144,000 | Yes. Asset, shaped to your workflow. |
The flip happens faster than most teams expect, and the table understates it: the custom system compounds as you extend it; the subscription resets to zero value the day you cancel. Build pricing detail: what a custom AI system costs in 2026.
The hybrid pattern that actually works
This is not a religious war. The teams getting real ROI run both:
- Buy individual productivity for everyone. Cheap, instant, low risk.
- Build the two or three workflows that define the business: the intake pipeline, the compliance engine, the pricing brain.
- Connect them. Built systems can call bought tools as components.
The mistake we see most: companies buy fifteen tools, build nothing, and wonder why nothing changed. Tool sprawl is not transformation. The P&L moves when the core workflow changes, and the core workflow is always custom.
Decide it in one meeting
Put the workflow on a whiteboard and ask five questions. Three or more "yes" answers means build:
- Does this workflow appear in our P&L or board deck by name?
- Does it require our proprietary data to work well?
- Does it need to take actions in our systems, not just answer questions?
- Would a competitor using the same SaaS tool erase the advantage?
- Are there compliance or audit requirements attached?
Still split? A one-week diagnostic settles it with numbers instead of opinions. That's how we start every engagement.
- Buy commodity workflows. Build differentiating ones.
- P&L rule: if the workflow has a name in your P&L, build it.
- 3-year SaaS spend often exceeds a custom build you'd own.
- Run hybrid: 80 percent bought, 20 percent built. The 20 makes the money.
Related questions
When should a company buy off-the-shelf AI instead of building?
Buy when the workflow is generic and the same for every company: transcription, meeting notes, email drafting, basic support deflection. Off-the-shelf wins on speed and price for commodity work, typically $20 to $100 per user per month with zero build time.
When is building a custom AI agent worth the cost?
Build when the workflow is a differentiator, touches proprietary data, needs to act inside your systems, or carries compliance obligations. If the workflow appears in your P&L by name, custom AI on it compounds; renting generic AI on it doesn't.
What is the hybrid approach to AI adoption?
Buy commodity tools for individual productivity, build custom systems for the two or three workflows that define your business, and connect them. Most mid-size companies should run 80 percent bought, 20 percent built.