Sherwood Callaway's pitch to overburdened engineering teams sounds almost heretical: forget metrics, forget traces—just send us your logs and let the machines figure out the rest.
It's a deliberately reductive thesis, the kind of contrarian stance that either sparks a category or crashes into the reality of production systems. Callaway, a two-time Y Combinator founder and former a16z Infrastructure Scout, is betting his newest venture—a San Francisco startup called Sazabi—on the idea that AI agents can synthesize the traditional three-pillar observability stack from log data alone. No separate instrumentation pipelines. No sprawling configuration files. Just logs in, intelligence out.
Sazabi emerged from stealth the week of April 8, 2026, according to coverage from SiliconANGLE, and the company is now running a closed alpha with design partners while targeting a broader platform launch later this year. Funding details remain undisclosed, though the startup confirmed it's raising seed capital with backing from more than 70 individual angels—a roster that reads like a who's who of developer tooling: employees from Vercel, Replit, LangChain, Anthropic, and Brex. Among them: Andrew Qu and Matthew Lenhard from Vercel, Matt Palmer from Replit, and Harrison Chase, the founder of LangChain.
That constellation of supporters signals something. Perhaps it's genuine belief in Callaway's vision. Or maybe just hedging—early bets on a founder with a track record, placed before the market decides whether logs-only observability is brilliance or folly.
A Brex Reunion, With Baggage
Callaway comes to this with some scars and some wins. His previous startup, Opkit, went through Y Combinator and was later acquired by 11x. Before that, he put in time at Brex and Crunchbase, roles that evidently left him skeptical of the complexity baked into modern observability tooling.
For Sazabi, he's assembled a team of 11, several of whom helped build the infrastructure and observability systems at Brex. The group includes Asif Arman, Andrew Aymeloglu, Ed Carrel, Alex Holovach, Justin Ko, Lewis Liu, Tom Nagengast, Rupa Vemulapalli, Henry Ventura, and Daniel Young, along with Hadley Callaway serving as Chief of Staff. It's a tight crew, the kind of ensemble that knows how monitoring breaks at scale because they've watched it break firsthand.
Sazabi went through Y Combinator's Spring 2026 batch and announced that affiliation in late March. The company even hosted a co-working session for fellow YC participants at Cursor's San Francisco headquarters—an alignment that feels intentional, given Cursor's prominence in the agentic coding movement.
The Product: Chat First, Config Never
At its core, Sazabi's proposition is radically simple. Instead of wiring up separate telemetry streams for metrics, traces, and logs—each with its own SDK, data model, and cost structure—teams just point their existing log endpoints at Sazabi. The platform ingests those logs, analyzes them in real time, and uses AI to synthesize the metrics and traces that engineers typically need. "Just change the endpoint and we'll be receiving logs," the company told SiliconANGLE.
The target customer isn't the platform engineer with a dedicated observability budget. It's the product engineer at a fast-moving startup, someone who wants to "instrument in minutes" without learning a query language or configuring alert thresholds. Sazabi accepts logs "in any format from any cloud/technology," per the company's website—a sweeping claim that will meet the test of messy, real-world log schemas soon enough.
Three features anchor the pitch:
Autonomous Alerts lean on AI to self-configure monitoring for error spikes, slow queries, failed deploys, and what Sazabi calls "frustrated users" and "runaway costs." Integrations include Slack, PagerDuty, incident.io, email, and webhooks—the usual suspects in incident response tooling.
Conversational Debugging is the chat-first interface: engineers ask natural-language questions, and the system generates visualizations and suggests fixes without requiring knowledge of Datadog's query syntax or PromQL. It's the kind of feature that sounds magical in demos and either delights or infuriates users in production, depending on how well the AI handles ambiguity.
Coding Agents Welcome is perhaps the most revealing positioning choice. Sazabi natively supports Claude Code, Cursor, and other IDE agents, plus an MCP server, REST and GraphQL APIs, and a CLI for log queries and instrumentation. The company is explicitly designing for a world where developers work through AI assistants as much as they work with them—a bet on the agentic coding trend that's either prescient or premature.
The homepage also mentions "Code Search" (linking logs to specific files, commits, and lines), "Dynamic Visualizations" (charts generated on the fly), and "Perfect Memory" (knowledge of past incidents). Security compliance badges for SOC 2, ISO 27001, and GDPR are displayed, though no audit reports are publicly linked—a detail that enterprise buyers will inevitably scrutinize.
Wading Into a Market in Motion
Timing, in startups, is half the game. Sazabi is launching into an observability market that's in the middle of an AI-native reinvention—everyone is shipping agentic features, often within weeks of each other.
Honeycomb launched "Agent Observability" in May. IBM announced evolved AI observability capabilities for Instana at its Think conference around the same time. OpenObserve introduced what it calls "Observability 3.0" with an autonomous AI SRE agent in late April. groundcover expanded AI observability for agentic workflows compatible with Google Vertex AI. Datadog's "State of AI Engineering 2026" report, released in April, underscored the operational challenges as AI workloads scale—a market signal that incumbents see the same shift Sazabi is chasing.
The open-source ecosystem is equally restless. ClickHouse acquired Langfuse, an open-source LLM observability tool, in January. LangChain's LangSmith markets itself as "AI agent & LLM observability." Arize Phoenix supports integrations across LangChain, LlamaIndex, DSPy, and Vercel AI SDK, with documentation updated through May.
In mid-March, Sazabi published a manifesto—titled, unsurprisingly, "The Sazabi Manifesto"—declaring traditional monitoring dead and staking a claim to chat-first, logs-only infrastructure. Manifestos are a startup rite of passage, a signal of conviction. Whether that conviction survives contact with customers moving from alpha to production is the question that matters.
What Happens Next
For now, Sazabi is accepting signups for its public waitlist. The company is hiring across security engineering, software engineering (agents, infrastructure, product), marketing, operations, sales, and customer support—standard roles for a seed-stage company preparing to scale. No design partners or paying customers have been named publicly.
Which leaves the core uncertainty unresolved: can AI agents really replace metrics and traces with log synthesis, or is this a clever demo that fractures under the weight of real-world production systems? Callaway has assembled a credible team, raised backing from insiders who've built developer tools at scale, and timed the launch to ride a wave of interest in agentic infrastructure.
But observability is littered with startups that promised to simplify the mess and instead added to it. The logs-only bet is bold. Whether it's also sound will depend less on the manifesto and more on whether engineers trust Sazabi's AI to wake them up at 3 a.m. for the right reasons—and only the right reasons.
