Build our own vertical copilot — or buy from a category vendor?
Local open-weight models eliminate monthly vendor fees, but usage-based billing for autonomous agents can trigger a 5x cost increase, making fragmented data a costly risk.
The question
We need a copilot for one of our core operational functions — sales, support, legal, or recruiting. The category has matured: vertical vendors now sell complete solutions at $30-100K/year. Build-from-scratch on our existing AI stack is feasible in 2-3 engineer-months. Do we buy the vertical product, build on our own stack, or hybrid (vendor + customization)?
Counsel's position
Build on our existing stack, leveraging internal AI capabilities to own the core workflow and data integration.
Verdict
The verdict: Build on our existing stack, leveraging internal AI capabilities to own the core workflow and data integration.
How the criteria decide
3 of 3 criteria resolved on cited evidence.
| Criterion | Favours | Evidence |
|---|---|---|
| vertical AI copilot vendors vs build-on-stack | Build on our stack | Local open-weight models eliminate monthly vendor fees and data transmission You can now run state-of-the-art models entirely offline, directly on your machine, with zero latency and absolute privacy. Artificial Intelligence in Plain English - Medium AI-assisted development makes building in-house a credible negotiation threat the build question is getting a whole lot easier and a whole lot less expensive. So I think it's opening up ways of thinking and pressure on existing vendors that they've never really had before. Vertical AI vendors win on deterministic workflows and domain data In regulated and enterprise markets, deterministic workflow plus a probabilistic model beats pure agentic, every time. |
| AI workflow tool lock-in and switching costs | Build on our stack | Usage-based billing for autonomous agents triggers a 5x cost increase When GitHub Copilot transitioned to strict usage-based billing (UBB) on June 1, 2026, it triggered a massive 5x cost increase in a single month. |
| category maturity assessment for AI tooling decisions | Build on our stack | Local open-weight models eliminate monthly vendor fees and data transmission You can now run state-of-the-art models entirely offline, directly on your machine, with zero latency and absolute privacy. Artificial Intelligence in Plain English - Medium AI-assisted development makes building in-house a credible negotiation threat the build question is getting a whole lot easier and a whole lot less expensive. So I think it's opening up ways of thinking and pressure on existing vendors that they've never really had before. |
Local open-weight models eliminate monthly vendor fees and data transmission
Given your existing AI stack, self-hosting smaller models guarantees proprietary data privacy while avoiding recurring cloud costs.
AI-assisted development makes building in-house a credible negotiation threat
Given the 2-3 month feasibility of building on your stack, you can use this capability to force better terms from vertical vendors.
Usage-based billing for autonomous agents triggers a 5x cost increase
If you choose a vendor or managed API, autonomous loops can quickly drain budgets compared to flat-rate human-in-the-loop tools.
Vertical AI vendors win on deterministic workflows and domain data
When evaluating the $30-100K/year vertical products, their value lies in the guardrails and data integrations, not the underlying LLM.
Bolting agents onto fragmented data causes agentic thrash and high costs
Building a copilot on top of a disconnected legacy stack will severely limit its efficacy and drive up token costs.
Read another verdict
- Which process should we point AI at first?
- Put one person in charge of AI — or is a Head of AI premature for us?
- Buy a tool for this process, or build around our own knowledge?
- Centralize AI strategy under CEO or distribute ownership?
- Adopt new AI ROI tools or refine existing methods?
- Invest in pre-build costing or post-deployment ROI tracking?
- Our documents are a mess. Clean them up before AI, or after?
- How do we measure the return on an AI workflow — and what baseline is honest?