RadiantOps disclosed Monday that it closed a $7 million seed round led by Foundation Capital, with Vermilion Cliffs Ventures, Essence, Vela and MKT1 joining the deal. The Durango, Colorado startup declined to share its valuation.
Its pitch centers on a gap that engineering managers have struggled to articulate: nobody really knows whether AI coding assistants make developers more productive or just faster at shipping problems. RadiantOps built software that traces which lines of code an AI agent wrote, tallies what each merge cost to run, and counts how many times human engineers had to circle back and fix the output.
The timing reflects a broader unease in engineering leadership. GitHub released Copilot Metrics as a general offering on February 27, 2026, but most platforms still boil developer activity down to percentile rankings—a method that can obscure whether an agent actually saved time or introduced the kind of technical debt that compounds quietly over quarters.
A Second Act
Travis Kimmel and Ben Thompson know this terrain. The pair co-founded GitPrime, which passed through Y Combinator's Winter 2016 cohort and sold to Pluralsight in 2019 for $170 million in cash. They launched RadiantOps in 2024 alongside Sigmon Myers, setting up in Durango rather than a coastal tech hub. TJ McNichols, who ran customer success at GitPrime and later held sales roles at Pluralsight, LaunchDarkly and Snowflake, signed on as vice president of sales.
The company opened demo requests on September 16, 2026, and declined to specify current headcount; LinkedIn showed a team of two to ten employees during the stealth phase in July 2026.
What Gets Measured
RadiantOps stitches together data from code repositories, project trackers, agent runs and collaboration tools into what it calls "episodes of work." The platform flags agent-authored changes, calculates spend per merge, and tracks rework rates. Sample metrics on the company's site include agent-authored merges, stalled changes, agent rework frequency, and time to first review.
Integrations span GitHub, Bitbucket, Jira, Linear, ClickUp, Slack and Microsoft Teams. On the agent side, it connects to Cursor, Codex, Bedrock and Databricks. The company said it holds SOC 2 Type II certification and supports SSO, SAML and read-only access.
Where the Money Goes
RadiantOps plans to expand its engineering team and get the product in front of more organizations. Foundation Capital's involvement suggests the firm sees a durable market in tooling that audits AI-generated work, though venture appetite for developer infrastructure has cooled considerably from the peak years of 2021 and 2022.
The round includes an interesting roster of backers. MKT1 appears on the firm's portfolio page; according to the firm's website, it continues supporting companies from its first fund while no longer making new investments. Vermilion Cliffs Ventures describes a focus on developer-led and "agent era" startups. Firsthand VC is listed as a backer on RadiantOps' homepage but does not appear in the formal press announcement, a discrepancy the company did not clarify.
The Market They're Entering
RadiantOps faces established players. Pluralsight Flow—formerly GitPrime, the very product Kimmel and Thompson built and sold—still tracks engineering metrics across Git and Jira. Harness Software Engineering Insights, assembled from the Propelo acquisition, added messaging in 2026 around tracking AI-generated code to production and measuring ROI on AI spend. Jellyfish released an AI Impact Platform and published benchmark research on AI's effect on engineering teams in March 2026. LinearB rolled out AI analytics features the same year.
The category has gotten crowded, but perhaps more crowded than clarifying. Many platforms offer dashboards; fewer promise to disentangle signal from noise when an agent ships fifty commits in an afternoon.
The Philosophical Case
Kimmel's framing resists the tidy narrative that more code equals more progress. "A team is not a percentile," he said in the announcement. "Knowledge lives in particular people and moves between them along paths no benchmark can see. A slow review might mean a design flaw is surfacing, or an engineer is exercising careful stewardship, or someone is on leave. The difference matters, and you cannot recover it by making a more confident guess."
Writing on the RadiantOps blog in July 2026, Kimmel argued that agents will eventually author nearly all code, shifting the bottleneck to deciding what should exist and what an organization can realistically maintain. "Traditional accounting often reads a day spent making those decisions as an empty day," he noted in press materials.
That worldview—where the hard work becomes editorial rather than generative—underpins the product. If AI writes the code, someone still has to decide whether it was worth writing. RadiantOps is betting that engineering leaders will pay for software that makes that judgment less guesswork and more arithmetic.
