Loadsmart is a next-generation freight technology platform that digitizes and optimizes transportation logistics for shippers, carriers, and warehouses.
To lead this platform into its next phase, CTO Ron Ben Yosef oversees an R&D organization of 120 people, closing in on 100 engineers. Loadsmart is pushing beyond its traditional platform toward a more agentic one, giving customers self-service tools to build or run their own AI agents, backed by a deployed engineering team for the workflows that need more customization. As Ben Yosef puts it, the goal is to stay “cutting edge” on AI without going to the extreme of chasing every new capability for its own sake.
Loadsmart has partnered with Jellyfish for more than five years, through a stretch of cost optimization in 2023 and early 2024, and now through a return to growth mode, leaning heavily on AI. Loadsmart uses Jellyfish to prove which AI investments actually work, govern how engineers use coding assistants and agents, and translate engineering signals and results into decisions the board can act on.
Proving Which AI Tools Actually Work
Loadsmart’s AI journey didn’t start with a clean win. About a year and a half ago, the company piloted GitHub Copilot, hoping coding assistants would move the needle.
pull quote
"We didn't see much productivity," Ben Yosef says. "I guess what we assumed is that, back then, it's more autocomplete capability that GitHub Copilot offered us, rather than what we see today with coding assistants." Jellyfish data showed cycle times holding flat or even slowing slightly, enough evidence for Loadsmart to pass on a full rollout.
Cursor told a different story, and it started from the bottom up. “Our engineers read about it, some of them started using it, playing with it, and saw a lot of value in it,” Ben Yosef says. “What happens is it becomes almost viral internally within the org.” Loadsmart formalized a pilot, measured it the same way with Jellyfish, and saw an immediate difference in cycle time and code review outcomes, enough to justify rolling Cursor out company-wide.
That rollout came with a side effect: coding assistants started producing roughly 2.5x more code than before, and code review couldn’t keep up. “This is where we’re using AI to help us with AI problems,” Ben Yosef says. Loadsmart brought in CodeRabbit for AI-assisted review and linked it into Jellyfish to measure the result. The review slowdown that had initially cost the team as much as 25% of its delivery time has now decreased to just 1–2%, an accepted tradeoff for overall cycle time improvements.
Governing Token Spend Without Slowing Anyone Down
After optimizing tool selection for their org, Loadsmart wanted to dive deeper into how specific teams were actually using AI. With the help of Jellyfish’s MCP, Loadsmart was able to measure model selection patterns and identify opportunities for workflow improvements.
“People would choose the strongest model to do a very simple, repetitive task,” Ben Yosef says. “Once we realized something was going on, this is where we can actually intervene, we need to train our org about best practices, we need to give some guidance, we need to set some guardrails.”
That visibility turned into actionable governance. Loadsmart implemented model-selection guidelines and provided developer training on prompt optimization, teaching engineers when to choose lighter models versus premium ones. This reined in token spend without asking anyone to sacrifice delivery speed.
Translating Technical Metrics to the Boardroom
At the board level, executive questions focus strictly on business impact. “We now generate 86% of our code via AI,” Ben Yosef says. “Okay, so what does that give us? That’s a very hard question to answer these days.”
Jellyfish helps answer that question of AI impact by showing whether the company is shipping more features, shipping them faster, and connecting those metrics back to sales results. “This is where Jellyfish really helps me… to understand what is the impact that we’re making, and why the ROI is there.”
The underlying shift, in Ben Yosef’s view, goes beyond any single metric: “The role of an engineer is changing these days. We’re no longer executing code, we’re more focused on planning, system thinking, and then reviewing the code after those coding assistants help us generate it.”
A Strategic Engineering Intelligence Platform
To manage this shift effectively, Loadsmart treats Jellyfish as a strategic engineering intelligence platform rather than a tracking monitor.
"I don't see Jellyfish as a reporting tool. Jellyfish is a decision making tool," Ben Yosef says. "It gives you all the information and insight you need to make a decision. If you want data driven decisions at our size, or bigger, you need Jellyfish to support that."
Ben Yosef’s advice to leaders earlier in their own AI transformation is direct:
"Any leader would be completely blind without understanding the baseline… how you're doing in terms of quality, productivity, and delivery… before introducing a transformation like this. Jellyfish gives you the ability to understand the before and after: how a change shapes a team, a group of teams, or an entire company. I think it's the only way to see that properly."
As the role of an engineer shifts and the SDLC becomes increasingly AI-native, Jellyfish gives Loadsmart the necessary intelligence to understand their agentic development lifecycle, measure true AI ROI, and continuously improve how their teams work with AI to generate results for the business.
Key Takeaways
Before Jellyfish
- 30-45% of time spent on maintenance: Leaders assumed support work took roughly 15% of team time, but with no way to catch it before it became a pattern, it often resulted in 30-45% of time spent on maintenance.
- 25% of human time spent on PR reviews: There was a 2.5x increase in PRs because of AI-assisted coding, but no reliable way to tell whether a new coding assistant was actually making teams faster. As a result, 25% of human time was being spent strictly on PR reviews.
- Untracked Token Spend: Engineers often defaulted to the strongest available model even for simple, repetitive tasks, with no visibility into the cost of that habit.
With Jellyfish
- ROI of 86% AI-generated code: An AI adoption index and delivery intelligence give the CTO a data-backed answer when the board asks what 86% AI-generated code is actually buying the business.
- Real-Time Allocation Visibility: One connected view across Jira, GitHub, and AI coding assistants lets managers catch misallocated time in weekly 1:1s and retrospectives instead of at quarter’s end.
- Evidence-Based Tool Selection: A measured GitHub Copilot pilot showed minimal gains; a measured Cursor pilot showed immediate ones. The data, not enthusiasm, decided the company-wide rollout.
- Governed AI Spend: Visibility into token usage and model selection let Loadsmart train engineers on best practices and set guardrails without slowing delivery down.
