As cloud undertaking monitoring software program monday.com’s engineering group scaled previous 500 builders, the group started to really feel the pressure of its personal success. Product strains had been multiplying, microservices proliferating, and code was flowing sooner than human reviewers might sustain. The corporate wanted a approach to evaluation 1000’s of pull requests every month with out drowning builders in tedium — or letting high quality slip.

That’s when Man Regev, VP of R&D and head of the Development and monday Dev groups, began experimenting with a brand new AI device from Qodo, an Israeli startup targeted on developer brokers. What started as a light-weight check quickly grew to become a essential a part of monday.com’s software program supply infrastructure, as a new case study launched by each Qodo and monday.com right now reveals.

“Qodo doesn’t really feel like simply one other device—it’s like including a brand new developer to the group who really learns how we work," Regev instructed VentureBeat in a current video name interview, including that it has "prevented over 800 points monthly from reaching manufacturing—a few of them might have induced severe safety vulnerabilities."

In contrast to code technology instruments like GitHub Copilot or Cursor, Qodo isn’t making an attempt to jot down new code. As a substitute, it makes a speciality of reviewing it — utilizing what it calls context engineering to know not simply what modified in a pull request, however why, the way it aligns with enterprise logic, and whether or not it follows inner greatest practices.

"You’ll be able to name Claude Code or Cursor and in 5 minutes get 1,000 strains of code," mentioned Itamar Friedman, co-founder and CEO of Qodo, in the identical video name interview as with Regev. "You might have 40 minutes, and you may't evaluation that. So that you want Qodo to really evaluation it.”

For monday.com, this functionality wasn’t simply useful — it was transformative.

Code Evaluation, at Scale

At any given time, monday.com’s builders are transport updates throughout lots of of repositories and providers. The engineering org works in tightly coordinated groups, every aligned with particular elements of the product: advertising, CRM, dev instruments, inner platforms, and extra.

That’s the place Qodo got here in. The corporate’s platform makes use of AI not simply to test for apparent bugs or fashion violations, however to judge whether or not a pull request follows team-specific conventions, architectural tips, and historic patterns.

It does this by studying from your personal codebase — coaching on earlier PRs, feedback, merges, and even Slack threads to know how your group works.

"The feedback Qodo offers aren’t generic—they replicate our values, our libraries, even our requirements for issues like function flags and privateness," Regev mentioned. "It’s context-aware in a method conventional instruments aren’t."

What “Context Engineering” Really Means

Qodo calls its secret sauce context engineering — a system-level method to managing all the pieces the mannequin sees when making a choice.

This consists of the PR code diff, after all, but in addition prior discussions, documentation, related recordsdata from the repo, even check outcomes and configuration information.

The concept is that language fashions don’t actually “assume” — they predict the subsequent token primarily based on the inputs they’re given. So the standard of their output relies upon virtually totally on the standard and construction of their inputs.

As Dana Nice, Qodo’s neighborhood supervisor, put it in a blog post: “You’re not simply writing prompts; you’re designing structured enter underneath a set token restrict. Each token is a design determination.”

This isn’t simply concept. In monday.com’s case, it meant Qodo might catch not solely the apparent bugs, however the delicate ones that sometimes slip previous human reviewers — hardcoded variables, lacking fallbacks, or violations of cross-team structure conventions.

One instance stood out. In a current PR, Qodo flagged a line that inadvertently uncovered a staging surroundings variable — one thing no human reviewer caught. Had it been merged, it might need induced issues in manufacturing.

"The hours we’d spend on fixing this safety leak and the authorized situation that it could deliver can be way more than the hours that we cut back from a pull-request," mentioned Regev.

Integration into the Pipeline

At present, Qodo is deeply built-in into monday.com’s growth workflow, analyzing pull requests and surfacing context-aware suggestions primarily based on prior group code opinions.

“It doesn’t really feel like simply one other device… It looks like one other teammate that joined the system — one who learns how we work," Regev famous.

Builders obtain ideas in the course of the evaluation course of and stay in command of last selections — a human-in-the-loop mannequin that was essential for adoption.

As a result of Qodo built-in instantly into GitHub by way of pull request actions and feedback, Monday.com’s infrastructure group didn’t face a steep studying curve.

“It’s only a GitHub motion,” mentioned Regev. “It creates a PR with the exams. It’s not like a separate device we needed to study.”

“The aim is to really assist the developer study the code, take possession, give suggestions to one another, and study from that and set up the requirements," added Friedman.

The Outcomes: Time Saved, Bugs Prevented

Since rolling out Qodo extra broadly, monday.com has seen measurable enhancements throughout a number of groups.

Inner evaluation exhibits that builders save roughly an hour per pull request on common. Multiply that throughout 1000’s of PRs monthly, and the financial savings shortly attain 1000’s of developer hours yearly.

These aren’t simply beauty points — many relate to enterprise logic, safety, or runtime stability. And since Qodo’s ideas replicate monday.com’s precise conventions, builders usually tend to act on them.

The system’s accuracy is rooted in its data-first design. Qodo trains on every firm’s personal codebase and historic information, adapting to totally different group types and practices. It doesn’t depend on one-size-fits-all guidelines or exterior datasets. All the things is tailor-made.

From Inner Device to Product Imaginative and prescient

Regev’s group was so impressed with Qodo’s affect that they’ve began planning deeper integrations between Qodo and Monday Dev, the developer-focused product line monday.com is constructing.

The imaginative and prescient is to create a workflow the place enterprise context — duties, tickets, buyer suggestions — flows instantly into the code evaluation layer. That method, reviewers can assess not simply whether or not the code “works,” however whether or not it solves the best drawback.

“Earlier than, we had linters, hazard guidelines, static evaluation… rule-based… you have to configure all the foundations," Regev mentioned. "Nevertheless it doesn’t know what you don’t know… Qodo… feels prefer it’s studying from our engineers.”

This aligns intently with Qodo’s personal roadmap. The corporate doesn’t simply evaluation code. It’s constructing a full platform of developer brokers — together with Qodo Gen for context-aware code technology, Qodo Merge for automated PR evaluation, and Qodo Cowl, a regression-testing agent that makes use of runtime validation to make sure check protection.

All of that is powered by Qodo’s personal infrastructure, together with its new open-source embedding mannequin, Qodo-Embed-1-1.5B, which outperformed choices from OpenAI and Salesforce on code retrieval benchmarks.

What’s Subsequent?

Qodo is now providing its platform underneath a freemium mannequin — free for people, discounted for startups by Google Cloud’s Perks program, and enterprise-grade for corporations that want SSO, air-gapped deployment, or superior controls.

The corporate is already working with groups at NVIDIA, Intuit, and different Fortune 500 corporations. And due to a current partnership with Google Cloud, Qodo’s fashions can be found instantly inside Vertex AI’s Mannequin Backyard, making it simpler to combine into enterprise pipelines.

"Context engines would be the massive story of 2026," Friedman mentioned. "Each enterprise might want to construct their very own second mind if they need AI that truly understands and helps them."

As AI programs develop into extra embedded in software program growth, instruments like Qodo are displaying how the best context — delivered on the proper second — can rework how groups construct, ship, and scale code throughout the enterprise.

Source link