SDE III - Engineering Productivity (AI)

Job description

Most boards and executives are currently flying blind when it comes to cyber risk. They are guessing. At Safe, we’ve built an AI-driven engine that finally gives the C-Suite a clear, quantified, and real-time view of their security posture. We don’t just provide data; we provide **certainty**. We are a **$170M Series C-funded** category leader. We don’t play in the mid-market; we operate at the highest levels of global enterprise. Today, we are proud to **serve 10% of the Fortune 500**, protecting global icons such as **Apple, Netflix, AT&T, Verizon, and Victoria’s Secret**. As we scale toward our next chapter, we are looking for high-performers who want to do the best work of their careers at the intersection of AI and Cybersecurity. ### [**The Culture Memo: Our Operating System**](https://jobs.safe.security/culture/) Safe is not a typical corporate environment. We are a high-intensity, mission-driven team. We value builders who want to define a category and work alongside people who are equally committed to excellence. - **Extreme Ownership:** We don’t do "not my job." We hire people who see a gap and own the solution from start to finish. - **The Elite Standard:** We serve the most sophisticated companies on the planet. Our work must be bulletproof. Whether it’s a line of code or a sales deck, we aim for Tier-1 quality every time. - **Methodology & Rigor:** We don’t wing it. From **Force Management** and **MEDDICC** in sales to **data-driven sprints** in engineering, we rely on proven frameworks to stay disciplined and predictable. - **Radical Candor:** We move too fast for politics or sugar-coating. We value direct, honest feedback that helps us find the right answer quickly. - **The Series C Hustle:** We have the stability of a well-funded leader but the heart of a startup. ### **The Perks & Ownership:** We want our team to feel like owners because they **are** owners. We trust our people to manage their results and their time. - **Meaningful Equity:** Every "Safestar" is a shareholder. You aren’t just an employee; you are a partner in our success. - **Unlimited Leaves:** We don’t believe in clock-watching. We offer **unlimited leave** because we trust you to take the time you need to recharge while staying committed to the mission. - **Comprehensive Benefits:** We provide top-tier medical insurance and wellness benefits to ensure you and your family are well cared for. - **Career Trajectory:** We are growing aggressively. For high-performers, the path for advancement moves at the speed of your ambition. This role exists to change how engineering at Safe gets work done — not to add AI tools on top of today's workflow, but to rebuild the engineering operating model around AI, automation, and redesigned process, and to prove the gain in numbers. You will look across the full lifecycle — requirements, design, implementation, review, testing, deployment, operations, incidents, documentation — find where engineers lose time, and decide what actually fixes it: an AI agent, an automation, a platform capability, or a process change. Then build it to production quality, drive adoption, and measure whether cycle time moved. This is not a Developer Experience, DevOps, or internal-tools role. It is a software engineering role whose product is engineering throughput. ### Why This Role Exists Engineering output is the constraint on how fast we ship against Fortune 500 demand. AI can lift it — but most organizations bolt AI onto unchanged workflows and get a chat window. The leverage is in redesigning the workflow around what AI can now do, putting agents inside CI, review, testing, and incident response, giving them real engineering context, and setting guardrails so engineers move fast without lowering the bar. The objective is not AI adoption. It is materially higher engineering output, speed, and quality. ### What You'll Do: - ind the bottlenecks: Instrument how engineering time is actually spent — cycle time, review latency, CI failures, test authoring, operational toil, onboarding — and rank problems by recoverable hours. - Pick the right instrument: Decide per bottleneck whether the answer is AI, automation, conventional software, or process change. Reject AI where it's the wrong tool. - Apply AI across the lifecycle: AI-assisted development and refactoring, first-pass code review, test generation and suite optimization, automated documentation and release notes, design analysis, automated migrations, AI-assisted debugging, incident investigation, and root-cause analysis. - Build engineering agents: Agents that diagnose CI failures, investigate production issues, write tests, run dependency upgrades, propose security fixes, analyze PRs, and execute repetitive migrations — human-in-the-loop by default, autonomy earned per workflow. - Redesign the process: Rework code review, testing, CI, and incident management for a world where AI does the first pass — and define what stays human-owned and where approval is mandatory. - Build the platform: AI gateway, agent runtime and workflow orchestration, codebase intelligence, engineering context retrieval, and deep Git/CI/observability/ticketing integrations — plus APIs and SDKs so teams build their own AI workflows on it. - Solve the context problem: Give agents permission-aware access to code, architecture docs, service ownership, deployment state, telemetry, incidents, and standards — and own retrieval quality and freshness. - Set the guardrails: Quality, security, privacy, access control, approval gates, auditability, and evaluation for AI-generated change. - Prove the impact and drive adoption: Baseline, experiment, publish results, kill what doesn't move the metric, and mentor teams into the patterns that work. **How We Measure Success:** Not by tools built, experiments run, LLM calls, or chatbot users. By outcomes: reduced cycle time and repetitive work, faster PR review and CI failure resolution, fewer escaped defects, faster incident diagnosis, less operational toil, shorter onboarding, higher deployment frequency and throughput, and real adoption of AI workflows. ### What We're Looking For: - 6+ years building production software, with SDE3-level ownership of systems in production. - Strong Python, Java, Go, or equivalent — you ship production services and review others' code. - Solid distributed systems, API and microservice, and software architecture grounding. - Hands-on cloud infrastructure and CI/CD experience. - Demonstrated developer tooling and automation work that other engineers actually used. - Practical experience with LLMs and agentic systems in real systems, not only experiments. - Ability to reason quantitatively about workflows: baseline, hypothesis, experiment, measured result. - Strong writing and the ability to influence engineering teams without authority. **Nice to Have** LLM APIs and model orchestration · production AI agents with tool use · RAG, retrieval, and vector databases · MCP or similar tool/context protocols · LLM evaluation, prompt and context engineering, AI observability · GitHub/GitLab APIs and deep CI/CD integration · Kubernetes and cloud platforms · internal developer platforms · DORA/SPACE-style engineering measurement · security or compliance-bound environments. **SDE3 Expectations** Not an execution-only role. You identify org-wide productivity problems without being handed a backlog, work across teams, read both technical and process bottlenecks, build for scale rather than prototypes, influence without authority, prove improvements quantitatively, and mentor engineers into AI-first ways of working. ### What Success Looks Like: - In 90 days: Engineering time-spend is baselined, top bottlenecks are quantified in recoverable hours, and the first automation is shipped and in use. - AI does the first pass: Code review, test generation, and CI failure diagnosis run through AI workflows by default; humans review judgment, not mechanics. - Cycle time drops measurably quarter over quarter, attributable to specific changes. - Agents carry real load across defined workflows, with approval gates, audit trails, and tracked reliability — while escaped defects and security findings do not rise.