Sajid Mehmood

Sajid Mehmood is the co-founder and CEO of Niteshift, a cloud platform that gives coding agents a running application and the tools to test their changes. A code diff can show what an agent changed; a browser check or an integration test can show whether the change fixed the problem. Sajid and co-founder Conor Branagan met at Datadog, where Sajid led teams that built tools for understanding how software behaves in production. At Niteshift, they apply that experience to the work engineers need to verify before it ships.

From network infrastructure to production diagnostics

Sajid earned a BSE in computer science from Princeton, graduating summa cum laude. With Curtis Huttenhower and Olga Troyanskaya, he co-authored Graphle, a tool for exploring dense biological networks. A researcher could select a gene and examine its immediate connections, then refine the view without loading the entire network.

Sajid worked on networking between Google’s data centers, led development on parts of Foursquare’s developer platform, and worked as an engineer at Segovia. Before joining Datadog, he was the founding CTO of Taptap Send, an app for sending money across borders. As a vice president of engineering at Datadog, he led teams that launched Serverless Monitoring, Network Performance Monitoring, and Cloud Cost Management.

Sajid also started and oversaw Datadog’s Applied AI group. The group trained the time series forecasting models Toto and Toto 2 and shipped Bits AI SRE. Datadog trained Toto on observability metrics alongside public time series data so it could forecast a new series without fitting a separate model to each metric. That matters for production data, where readings can arrive seconds apart, contain gaps, or spike during incidents. Bits AI SRE uses telemetry to investigate possible causes of a production problem. It shows responders the evidence behind its findings and distinguishes supported explanations from ones the available data leaves unresolved.

What an agent needs beyond the repository

When Sajid and Conor tried Claude Code, they expected coding agents to help them ship faster. As the agents took on more work, the founders kept running into the same bottleneck: an agent could make a plausible change in a cloned repository and still leave an engineer to start the application, reproduce the bug, and check the fix. Running several complete development environments on a laptop also limits how many tasks can proceed at once. Niteshift’s design addresses both problems by running full application environments in parallel in the cloud.

• Full-stack verification. A setup agent configures the services a task needs, including authentication and seeded test data. With the application running, a coding agent can reproduce a bug in a browser, test its fix, and return a pull request with evidence a reviewer can inspect. At Listen Labs, engineers point agents at evaluation scripts so they can try changes to quote extraction and measure whether accuracy improves across experiments.
• One environment, multiple agents. Teams configure their application, tools, and policies separately from the agent that uses them. An engineer can switch from Claude Code to Codex or another supported agent while keeping the same services and checks.

Sajid applies the same emphasis on outcomes to how teams measure agent adoption. He proposes tracking the share of shipped software that agents produce autonomously. Token usage measures how much work the agents attempt; the share that ships measures how much makes it through the team’s checks and review.