Highlights
- —Built a reproducible organization that no longer depends on specific individuals
- —Resolution of Salesforce at a level that general-purpose AI can't reach
- —Re-directed freed-up capacity from maintenance into offensive work
About LegalOn Technologies
— Tell us about LegalOn Technologies.
Shimoda: We're a legal-tech company developing and delivering AI-powered legal workflow solutions. As of the end of September 2025 we serve more than 7,500 customers globally, and we're still expanding.
The two of us are in the Business Operations Section, which maintains, manages, and builds on Salesforce as well as the tools used by sales and marketing. The section has a development team and a planning team that handles requirements and operational design. We respond to requests from many teams across the company while doing Salesforce-centric development.
Aiming for a reproducible organization that uses AI to scale beyond head-count
— What led you to adopt Blaze?
Kataoka: From late 2024 we had been wanting to use AI agents to streamline and standardize our development process. We tried connecting a general-purpose AI agent with GitHub and Slack and built some in-house automation for requirements writing. More recently we were experimenting with the MCP server Salesforce released, looking for ways to increase efficiency and reduce variation.
But building things from the ground up like that is genuinely hard. While we were exploring, Blaze was introduced to us as a solution that fit our needs almost perfectly. When we actually used it, we found that not only requirements drafting but also release work became faster, and we decided to adopt it.
— What started your AI exploration in the first place?
Shimoda: As the organization grew, we felt that "continuing to hire more people" wasn't realistic. We wanted to build an organization where we could keep enhancing things without being dependent on headcount. We were still hiring, but the market for the kind of talent we needed made it hard. We wanted a reproducible organization even with limited resources.
Kataoka: We do Salesforce development and operations fully in-house and release every week. With our current team alone there simply isn't enough development capacity, so we tap into the planning team's time as well and run the entire section together. In that structure, differences in individual knowledge and skill become visible, and it's hard to keep track of what everyone is building. Beyond pure efficiency, standardizing the process to level quality was a major challenge.
The deciding factor: Blaze's depth of Salesforce understanding
— What sealed your decision?
Kataoka: The depth of the product's Salesforce understanding. It smoothly covered everything from development to deployment. Even without fine-grained prompting, the first answer usually meets 80–90% of expectations. It genuinely felt like adding a new engineer to the team.
Shimoda: Getting strong answers without having to heavily tune system prompts matters enormously in real work. For impact analysis Blaze inspects Salesforce at the metadata level, so output quality is very high. If you think of the cost of hiring one more engineer, giving everyone on the team Blaze raises the overall quality of output far more.
— Any specific advantage over other AI tools?
Kataoka: We previously used Devin, which is a great tool, but rolling it out to the whole team had hurdles. Devin isn't specialized for Salesforce, so to get it to a usable level you have to heavily shape the context around Salesforce, and the initial setup and ongoing maintenance are costly. Blaze is Salesforce-native from the start, so we barely have to maintain that layer. After rollout, anyone can just state what they want to do in natural language and get the expected result. Lowering the "development bar" and letting everyone standardize around it — that immediate usefulness fit our team well.
Another thing worth noting is that Blaze is low on hallucinations compared to general AI — where you'd usually wonder, "Where did it pull that information from?"
"It flipped our assumptions" — what impressed the engineers
— Anything that specifically impressed you once you started using it?
Kataoka: The accuracy and speed of source-code grep across the codebase. Initially we assumed you couldn't meaningfully scan that much source locally. In practice it was faster and more accurate than we expected. That experience flipped our assumptions about what AI could do with information retrieval.
Shimoda: The most impressive moment for me was watching it build a Salesforce Flow from natural language. Flows aren't code — you build them visually in the UI. That you can describe one in plain language and have Blaze assemble it was something I honestly doubted until I saw it work.
Kataoka: Flows have a unique structure, so I thought they'd be hard for AI to handle. What makes Blaze special is that even when a deployment fails, it retries on its own. It diagnoses the error, fixes it, and keeps going until deployment actually succeeds. No unnecessary "I give up" moments or bouncing back to the human. Watching it move autonomously was genuinely moving.
A partner for everything from small releases to project brainstorming and report building
— Where do you typically use Blaze?
Kataoka: Right now Shimoda and I are the ones using it. I use it in three main ways. First, for weekly release tasks — small things like minor flow tweaks or updating picklist values. I hand the requirements to Blaze and it handles everything through deployment. Honestly even manually it would only take a few minutes, but I don't want to spend my thinking time on those small tasks, so I delegate aggressively.
I also use it on larger projects that span two or three weeks. I break down tasks, brainstorm with Blaze, and have it write code. And for standard reports — when I need a temporary report in a development org, I can just say "make this" and it's ready right away, which is incredibly useful.
— How do you use the brainstorming specifically?
Kataoka: This is my personal take, but when using AI, rather than using it to skip understanding, I find using it to deepen understanding leads to more efficiency overall. So I phrase my questions like "I'm thinking about it this way — is there a better approach?" Cross-checking my reasoning with the AI's response prevents misalignment in the actual development.
— Any other places where you've applied Blaze?
Shimoda: Recently our PMM (product marketing manager) tried it for pricing strategy. "If we change the current pricing, what's the impact?" — aggregating and drawing insights. In the past we exported data to spreadsheets and built complex formulas, but when we asked Blaze directly things like "How many records match these criteria?" and "What's the impact of the change?", it autonomously queried opportunity and contract data in the CRM and delivered precise insights. Blaze is a tool everyone on the business side can use, not just developers.
Turning the capacity Blaze creates from "maintenance" into "offense"
— What's changed in your work or your team since adopting Blaze?
Kataoka: I no longer have small weekly tasks eating my time, so I can concentrate on environment improvements and bigger projects I'd wanted to tackle. The feeling of "I can't start because I don't have time" is completely gone.
Shimoda: The biggest shift is that we can hand maintenance-level tasks to AI and redirect people to releases that have real business impact. Blaze can also answer day-to-day internal questions like "how does this spec work?" instantly. Communication overhead has dropped, and we can focus on our core work. That ability to create capacity is the biggest and most fundamental effect.
Just go touch Blaze
— Any closing message for companies considering adopting Blaze?
Kataoka: AI agents for Salesforce development are still a blue ocean, and Blaze is at the leading edge while being both high-quality and easy to adopt. Just look at the accuracy of its test-class generation — you'll immediately see how deeply it resolves Salesforce. I'd strongly recommend it to any company struggling with development, operations, or maintenance.
Shimoda: Honestly, just go touch Blaze. Have one person try it out — fetch metadata and have it do impact analysis, or have it build a flow in natural language. Once you've touched it, you'll immediately understand why we're this excited. I'm confident you'll want to roll it out to everyone.
——
Interview conducted in January 2026.
More about our products
