Building an AI-Native Team

1. 1
   Talk18 min

   Agentic SDLC: Building Blocks for Uber's Software Factory

   Uber

   Model gateways, agent identity, PII redaction, MCP gateways, tool access, remote environments, context graphs, and pre-CI validation.

2. 2
   MCP portals18 min

   Gateways Are All You Need

   Anthropic's Karan Sampath

   And Tobin South, What Does Enterprise-Ready MCP Mean? (14 min) — identity, access policy, observability, provisioning, oversight, and data-loss prevention.

3. 3
   Architecture25 min

   Scaling MCP adoption: reference architecture for enterprise deployments

   Cloudflare

   .

4. 4
   Authorization

   Enterprise-Managed Authorization

   MCP

   And the technical specification (20 min).

5. 5
   Evidence on adoption60 min

   2025 State of AI-assisted Software Development

   DORA

   AI amplifies the strengths and weaknesses of the surrounding engineering system. Pair it with the counter-evidence: METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (20 min).

6. 6
   Video42 min

   The Future of Software Creation

   Amjad Masad of Replit

   Agent infrastructure, secure execution, autonomy levels, and how cheap software creation may reshape teams.

• Link

  LiteLLM AI Gateway

  Gateways for the build

  (open source) and Cloudflare AI Gateway. Routing primer: Vercel, Six LLM routing strategies.

• Link

  MCP Registry

  Internal registries: the

  .

• Link

  AI Capabilities Model

  Adoption evidence: DORA

  ; Stack Overflow, 2025 Developer Survey: AI; OpenAI, How OpenAI uses Codex (PDF); Anthropic, How Anthropic teams use Claude Code.

• Adoption playbook18 min

  Building an Autonomous Engineering Org

  A maturity model covering champions, repository readiness, delegation, review bottlenecks, isolation, and organizational impact.

• Link

  Accelerate

  Nicole Forsgren, Jez Humble, and Gene Kim

  Useful for evaluating whether AI adoption improves delivery rather than merely increasing output volume.

Put your model calls behind one gateway or proxy (LiteLLM or Cloudflare AI Gateway). Add per-agent identity, budgets, model routing, fallback behavior, cost and latency logs, and redaction rules. Create a small MCP portal with allowlisted tools, explicit scopes, audit logs, and revocable credentials.

Write a one-page team adoption policy covering approved data, mandatory human decisions, incident ownership, review requirements, and what metrics matter. Prefer change-failure rate, lead time, review burden, cost, and verified task success over lines of code.