AI agents are becoming capable of investigating software incidents, but their effectiveness still depends on the telemetry available to them. bitdrift, the mobile observability company spun out of Lyft, is targeting that limitation with bitdrift AI, a platform designed to give AI agents real-time access to unsampled mobile telemetry and let them investigate and respond to problems without waiting for a new app release.
The next challenge for agentic AI in software engineering may not be the intelligence of the models. It may be the quality of the data they can access.
That is the problem bitdrift is trying to address with bitdrift AI, a new mobile observability system that gives AI agents programmatic access to real-time, full-resolution telemetry from mobile applications.
The company, which spun out of Lyft and says its technology has been installed more than a billion times across hundreds of millions of devices, is positioning the product as infrastructure for autonomous software troubleshooting.
The distinction is important. Traditional observability platforms commonly rely on sampling to control the enormous volume of telemetry produced by modern applications. Sampling makes large-scale monitoring economically practical, but it can also leave gaps precisely when an engineering team needs detailed evidence about an individual user’s experience.
Mobile applications present an even harder observability problem. Network conditions vary dramatically. Devices have different hardware and operating-system configurations. Users interact with applications in unpredictable ways. A problem that affects a small cohort can therefore be difficult to reproduce in a controlled engineering environment.
bitdrift AI is designed around full-resolution mobile telemetry instead.
The platform captures logs, traces and session context on the device and stores them in an unsampled ring buffer before streaming telemetry to its control plane. Engineers and AI agents can then retrieve targeted sessions and investigate specific cohorts without requiring the application to ship a new version.
That last capability is particularly relevant to the emerging market for AI coding agents and autonomous software engineering.
Systems from Microsoft, GitHub, Google, Amazon and other vendors can increasingly write code, inspect repositories and propose fixes. But an agent investigating a production mobile incident still needs reliable evidence about what users actually experienced. A model working from incomplete logs can produce a technically plausible explanation that has little connection to the underlying problem.
bitdrift’s approach is to move more of that evidence directly into the agent’s workflow.
The platform exposes workflows, charts, issues and captured sessions through a command-line interface, public API and customizable Skills. According to the company, agents can query telemetry across millions of devices, create instrumentation dynamically and iterate on investigations without waiting for another mobile application release.
This creates a feedback loop that looks different from conventional observability: detect an anomaly, gather targeted telemetry, investigate the affected users, test an explanation and potentially move toward remediation without treating every diagnostic step as a new development cycle.
The architecture also uses targeted data collection to avoid overwhelming an AI model with irrelevant information. Local capture buffers and server-controlled targeting allow an agent to request context from particular devices, sessions or cohorts instead of processing an undifferentiated stream of mobile events.
That matters because AI agents have finite context windows and engineering investigations can generate enormous amounts of telemetry. More data is not automatically better if the agent cannot identify which observations matter.
The company’s pitch is therefore not simply “more observability.” It is observability designed for machine-driven investigation.
ThredUp’s mobile engineering team is among the early users. Valerii Kuznietsov, the company’s senior staff software engineer and mobile engineering lead, said bitdrift has allowed the team to address a much larger share of customers experiencing minor issues. The company claims beta users of bitdrift AI have seen a 10X improvement in mean time to resolution, although those figures are vendor-reported and should be evaluated against independent production benchmarks.
For enterprise engineering teams, the potential benefit is substantial if the technology performs as advertised. Mobile incidents are expensive not only because of downtime or degraded experiences, but because engineers can spend hours reproducing problems that affect only a small percentage of users.
An AI agent with access to production-grade session data could potentially narrow that search dramatically.
The competitive landscape, however, is crowded. Platforms including Datadog, New Relic, Dynatrace and Splunk have built extensive observability capabilities spanning infrastructure, applications and increasingly AI-assisted operations. Cloud providers such as Amazon Web Services, Microsoft Azure and Google Cloud also offer telemetry, monitoring and machine-learning infrastructure.
bitdrift’s differentiation is narrower: mobile-first observability combined with unsampled telemetry and an interface explicitly designed for AI agents.
That specialization could become valuable as software teams adopt agentic development workflows. It could also become a constraint if enterprises prefer consolidated observability platforms that cover mobile, web, backend infrastructure and security in a single environment.
The broader trend is nevertheless clear. Observability is moving from dashboards designed primarily for humans toward systems that can feed automated reasoning and remediation.
For mobile engineering organizations, the important question will be whether AI agents can move beyond explaining incidents to safely taking action. That requires reliable telemetry, granular permissions, audit trails, human controls and integrations with deployment pipelines.
bitdrift AI addresses the data layer of that equation. Whether it can become a foundation for autonomous mobile operations will depend on how accurately agents can turn that additional visibility into fixes—and how confidently enterprises are willing to let them act.
Market Landscape
The observability market is entering an AI-driven transition. Traditional platforms were built around engineers querying dashboards, alerts and traces. Newer systems are increasingly being designed so AI agents can query telemetry programmatically, correlate events and execute remediation workflows.
bitdrift is taking a specialized approach focused on mobile applications, where intermittent connectivity, device fragmentation and difficult-to-reproduce user experiences make observability particularly challenging.
The larger opportunity is agentic observability: giving software agents enough production context to diagnose incidents and potentially resolve them without human intervention.
This sits alongside the broader growth of AI-assisted software development. Gartner has predicted that by 2028, 75% of enterprise software engineers will use AI code assistants, up from less than 10% in early 2023. That evolution increases the importance of reliable production feedback loops because generating code faster does not necessarily mean validating software faster.
For enterprise teams evaluating solutions, the key comparison points should include telemetry completeness, mobile coverage, data retention, AI-agent APIs, security controls, integration with CI/CD systems, and whether automated actions can be governed and audited.
Top Insights
- bitdrift AI gives software agents unsampled mobile telemetry, enabling faster investigation of user-specific problems without waiting for new application releases or SDK updates.
- The platform targets a central weakness in AI-assisted engineering: agents need high-quality production evidence before they can reliably diagnose and resolve software incidents.
- On-device ring buffers and targeted data collection let agents retrieve relevant sessions and cohorts while limiting unnecessary telemetry that could consume valuable model context.
- bitdrift’s mobile-first strategy differentiates it from broader observability platforms, while enterprises must weigh specialization against consolidated tools from Datadog, Dynatrace and cloud providers.
- The launch signals a broader shift toward agentic observability, where AI systems move from monitoring dashboards toward autonomous investigation, remediation and deployment feedback loops.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
