Situation
While building the agentic analytics platform, I pointed it at engineering attrition, onboarding and early productivity, which had never been measured.
Objective
Put hard numbers on the cost of developer onboarding, productivity loss and turnover, make the case for a formal Developer Experience organization, and give leadership a decision-ready path for the AI tooling that organization would need.
Approach
The DevEx business case
I finished the analysis and built a business case for the new CTO, framed around the President’s mandate to accelerate the innovation engine. The deck measured time to first contribution, time to median productivity, early termination and their annual cost.
The ask was a Sr. Director-led DevEx org, a small founding team reporting to the President of Product & Technology, for a modest fraction of what the problem already cost. It was projected to pay for itself many times over in year one.
Four selection initiatives
DevEx at SentinelOne covered developer tooling: a custom port of Backstage and Kubernetes deployment automation via Argo. I was given four concurrent initiatives, each running investigation, POC, selection and deployment:
- P1, AI coding assistants.
- P2, AI PR review agents.
- P3, SEI platform, for Software Engineering Intelligence, the DevEx reporting category.
- P4, LLM gateway, added later.
A repeatable selection practice
I used AI assistance to document the process I wanted to follow, modeled in part on SentinelOne’s own procurement process: requirement collection, product identification, assessment and selection, POC, vendor onboarding, internal reviews, governance, compliance, cohort selection, cohort-onboarding automation and stakeholder reporting. From that, the AI generated a repeatable practice for each initiative, including requirements inferred from documentation of current practices and the Confluence content for each project.
Running the program
I ran the initiatives with zero dedicated headcount and no budget authority. For each I produced evaluation criteria, vendor questionnaires, POC execution guides, assessment reports and comparison matrices, published as 22 Confluence pages across two spaces and as vendor-ready PDFs in Google Drive. I ran the vendor-security pipeline for 12+ vendors myself: Graphite Connect onboarding with its automatic NDA, ARB submissions and VRA processing. I coordinated directly with InfoSec, Legal and Compliance and was the sole internal point of contact in each vendor’s shared Slack channel.
For P1 I evaluated six AI coding tools against a weighted rubric of pricing, IDE and language coverage, code quality and security, developer experience, and enterprise readiness. The decisive finding was architectural: editors built as forks of VS Code cannot load Microsoft’s C++ extension, which ruled them out for the ~60-engineer Windows C++ team. The formal recommendation was a two-tool architecture: GitHub Copilot Enterprise, an extension-based assistant, as the primary IDE tool, plus Claude Code, a CLI agent, for async and reasoning work.
The AI coding POC
I took AI coding all the way to a live, month-long structured POC with a large engineering cohort from the Product & Technology org, under a data-governance approval I obtained from the Chief Trust Officer. I built the full cohort lifecycle automation, from manager-submitted intake through candidate enrichment and selection to Okta provisioning, with no IT involvement required. I wrote the getting-started guide, intake forms, and mid- and post-POC surveys.
Net results
- A measured, board-ready business case for a formal DevEx organization, covering time to first contribution, time to median productivity, early termination, and the annual cost of productivity loss and turnover.
- Four concurrent tool selections run through one repeatable practice, with 22 published Confluence pages and 12+ vendors taken through the security pipeline, with zero dedicated headcount.
- A formal two-tool recommendation for AI coding, driven by a compatibility finding for the ~60-engineer Windows C++ team.
- A live AI coding POC in March 2026 with automated cohort provisioning and a data-governance approval in place.
Takeaway: run a selection like a system. Modeling the whole procurement-to-cohort lifecycle once, then generating each initiative from it, is what made four concurrent evaluations workable solo.
Scope and technologies
AI-assisted process authoring; SentinelOne procurement workflow; Confluence and Google Drive; the vendor-security pipeline of Graphite Connect with automatic NDA, ARB and VRA; Okta provisioning; Slack vendor channels; Backstage as a custom port; Argo for Kubernetes deployment automation. Categories evaluated: AI coding assistants, AI PR review, SEI reporting and LLM gateways.