The sales team wants AI Agent to help prepare quotes, and IT starts evaluating tools. But management raises a question first: Our pricing rules reflect years of company experience—should we build the entire system ourselves? If your team is stuck on the same question, we recommend separating two issues: which practices the company must control, and which software it needs to develop itself. The two are related, but they do not have to be decided together.
The introduction to a Latent Space interview notes that Monaco builds its own AI Agent workspaces—the environments in which AI Agent performs tasks—while purchasing code review tools to check code changes. Building and buying can coexist, but the episode introduction does not explain why the team made those choices. The recommendations below are therefore the Intellicon Solutions FDE team's guidance for enterprise evaluations, not an account of the interviewed team's decision-making.
Break Down the Quoting Workflow Before Deciding What to Build Yourself
Consider a quoting workflow that a company plans to support with AI Agent. Sales needs to organize customer requirements, look up pricing data, apply discount rules, and ask a manager to approve the quote. These steps all contribute to the same quote, but each requires different data and tools. Deciding at the outset to build everything or buy everything prevents the team from comparing options step by step and finding the right approach for each.
We recommend that sales managers and IT jointly document the data each step needs, the output it must deliver, any company-specific rules, and who will maintain it over time. For example, which documents should be reviewed when gathering customer requirements, and who should be asked if a delivery date is missing? Which policy governs discounts, and who decides on exceptions? The team needs clear answers before it can determine whether a tool is sufficient.
Next, ask the team to place each step into one of three categories:
- Existing tools can handle it: The tool can already process the required data and produce the necessary output, but a hands-on trial is still needed to confirm that the results meet operational requirements.
- The company needs to configure its own rules and workflows: The tool already provides the execution capabilities, while the company needs to configure discount conditions, document formats, and management approval steps.
- Existing capabilities are genuinely insufficient, so custom development warrants evaluation: If the team has tested existing capabilities and adjusted the settings but still cannot meet essential requirements, compare custom development with purchasing another tool for that specific step.
Once the list is ready, ask the sales manager to confirm which requirements are essential, while IT identifies the available tools and the questions that still need testing for each item. A step's importance does not automatically mean it requires custom development.
Owning Your Pricing Rules Does Not Mean Building Every Tool
Suppose the company offers different discounts to different customers, and managers require additional approval for special payment arrangements. The company needs to define, review, and update these rules. But the team can first evaluate whether existing tools can be configured to execute them, rather than immediately developing a system from scratch.
EgentWrX, for example, lets companies create their own skills to capture practices and rules that need to be applied consistently. The platform also provides repeatable tasks that can be linked into workflows, where work passes to the next step after the previous one is complete. A company can start by documenting its quote-checking practices as a skill, then configure recurring steps such as organizing requirements and preparing quotes as tasks, and test whether they meet its needs.
For situations where a quote exceeds a department's limit or its terms are unclear, the team should first designate who is responsible for approval. It should then configure the workflow to require human approval before proceeding to the next step, pausing the process until that approval is given. Testing should verify not only whether AI Agent produces a quote, but also whether the company can clearly configure the sequence of work and assign responsibility for every step that requires review.
Even when using existing tools, the company cannot leave decisions about rule maintenance to the vendor. We recommend having the sales manager validate the rules, with IT responsible for configuration and integration. Both sides should then agree on how changes will be tested. If configuration changes still cannot meet essential requirements, document the specific limitations and add them to the list of issues to resolve.
Whether You Build or Buy, Run Hands-On Trials and Assess the Maintenance Effort
When selecting tools, we recommend asking the team to test a few representative work samples: a quote with complete information, an inquiry missing required fields, and a case with special conditions that require management approval. These are test scenarios, not examples of proven implementation results. The purpose of the trial is to determine whether each step can be completed and who can address problems when they occur.
When documenting issues, identify the exact step where the process breaks down. For example, will AI Agent continue preparing a quote when the customer has not provided a delivery date? After discount rules change, does the person responsible know what to update and which samples need to be retested? The team should first establish whether the issue stems from incomplete data, unclear rules, or missing tool capabilities, then decide whether to supply more data, adjust the configuration, or develop new functionality.
Ongoing maintenance also needs to be part of the comparison. For custom development, establish who will handle testing, bug fixes, and continued updates. For purchased tools, clarify which issues the vendor will address, which settings the company must maintain itself, and how the company can retrieve the data it needs if it stops using the tool. Do not assume that buying is always cheaper. Nor should having engineers on staff lead you to overlook the ongoing staffing commitment required to maintain internally developed software.
Before the next budget discussion, ask sales and IT to prepare a workflow checklist, trial records, and a division of maintenance responsibilities, then meet to review them. If existing tools combined with company-configured rules and workflows can meet essential requirements, you can adopt that approach and continue validating it through testing. If a missing capability genuinely affects essential work, and the company has someone who can maintain it over the long term, evaluate custom development for that specific step. There is no need to rebuild other tools that already work.
FAQ
Should all core workflows be built in-house?
Not necessarily. Start by identifying which rules the company must control, then determine whether the software needs to be developed in-house. In quoting, for example, the company should decide discount conditions and how to handle exceptions. But the team can first configure and test those rules in existing tools, and consider custom development only after confirming that the tools cannot meet essential functional requirements.
What is the company still responsible for after buying a tool?
The company still needs to define requirements, update rules, and verify results. It cannot hand all of these responsibilities to the vendor. We recommend that sales managers and IT jointly assign responsibility for ongoing maintenance. Before purchasing, they should also clarify which issues the vendor will address and how the company will retest after changing its workflows.
How do we determine whether a missing capability is worth building ourselves?
First, confirm that the missing capability affects essential work and cannot be addressed by completing the data or adjusting the configuration. Then evaluate custom development. Ask the team to provide trial samples, document the expected results and actual problems, and assign someone to handle ongoing maintenance. This helps avoid investing in development before the requirements have been validated.
References
- How Stanford Teaches AI-Native Engineering — Mihail Eric, Monaco — Latent Space, published November 8, 2026. The episode introduction notes that the interviewed team combines internally built workspaces with purchased code review tools, without explaining the reasons for those choices. This article uses that observation as a starting point for enterprise evaluation recommendations.



