§ 01 — OPERATING THESIS
Developer Experience is the product. Engineers are the customers.
Forsgren & Storey's research identifies three orthogonal levers for engineering productivity. A platform that fails to move all three is a platform that has shipped tools, not outcomes.
PILLAR · 01
Flow
Time spent in deep, productive work. Reduced context-switching, fewer interruptions, golden paths that remove decision fatigue.
Lever: Backstage scaffolders, opinionated templates, self-service provisioning.
PILLAR · 02
Feedback Loops
Time from change to signal. Faster builds, faster tests, faster production telemetry — and faster loops between humans and the systems they operate.
Lever: CI/CD throughput, pre-prod parity, contextual telemetry.
PILLAR · 03
Cognitive Load
Surface area an engineer must hold in working memory. The platform absorbs incidental complexity so teams can spend their thinking on the domain.
Lever: Crossplane abstractions, paved roads, intent-based APIs.
§ 02 — THE NEXT LAYER
Context is the competitive moat.
In 2026 every engineering org has access to the same models. The differentiator is no longer compute — it is the structured meaning of the signals fed to those models. Context engineering is to the AI-native enterprise what DevOps was to the cloud-native enterprise: a capability that compounds.
Active telemetry
Signals that self-describe — carrying ownership, intent, and lineage at the point of origin. Enrichment at the edge, not the warehouse.
Context graphs
Live topology of services, deployments, and causal edges. The substrate that lets both humans and agents answer why, not just what.
Purposeful pipelines
Telemetry shaped before it moves. Adaptive sampling, intent-aware routing. Cost-per-insight replaces cost-per-data-unit.
Two companies use the same model. One feeds it raw logs; the other feeds it structured, semantically coherent, domain-aware signals. Only the second achieves real intelligence — and the gap compounds.
Working principle · context engineering
§ 03 — MEASUREMENT
DORA + SPACE on a single vendor-agnostic pipeline.
Five DORA metrics for delivery health. Five SPACE dimensions for human signal. One pipeline — OpenTelemetry and CDEvents — so the platform's own measurement is portable, queryable, and ready to feed both dashboards and reasoning agents.
EMITCDEvents from CI/CD
OTel from runtime
→
ENRICHService catalog
Ownership · intent
→
ROUTEOTel Collector
Adaptive sampling
→
SERVEDashboards
Agents · SLOs
DORA: Deployment Frequency · Lead Time · Change Failure Rate · MTTR · Reliability · SPACE: Satisfaction · Performance · Activity · Communication · Efficiency
§ 04 — ADOPTION ROADMAP
Phased delivery, measured at every step.
Not a big-bang transformation. Each phase ships a measurable capability, earns adoption on its merits, and creates the substrate for the next.
0
Backstage stand-up, service catalog seed, platform Net Promoter Score (NPS) baseline, OTel collector deployed.
Time-to-first-PR for new hires
1
Scaffolder templates for top 3 service archetypes. CDEvents wired into CI/CD. First DORA dashboard.
% services on paved road
2
Crossplane Compositions for RDS, S3, Secrets. Self-service provisioning behind RBAC. Platform NPS > +30.
Lead time for new infra
3
INFLECTION
Durable Workflows
Q3
The platform graduates from stateless services to durable workflows. Temporal worker template, signed-execution patterns, audit trail wired to context graph.
Workflow reliability · audit coverage
4
Advisory agents on platform telemetry. Compliance attestation as Temporal workflow. Context graph feeds AI Factory upstream.
Cost-per-insight · agent correctness
§ 05 — QUICK WINS & AI-AUGMENTED DEVEX
Adoption is earned in the first 90 days.
Two parallel tracks: visible developer wins that build the platform's reputation, and a deliberate AI tooling strategy that treats context as the input — not the model.
Quick Wins · First 90 Days
- One-click service: form-to-running on EKS in ≤10 min via Backstage scaffolder.
- NPS survey: baseline + qualitative follow-up. Quarterly cadence.
- Deployment dashboard: DORA snapshot per team, no setup required.
- Secrets self-service: External Secrets + RBAC, retire ticket queue.
- Office hours: weekly platform clinic. Adoption is a conversation.
AI Tooling · Copilot Suite + Claude Code
- Copilot Suite: default IDE assist for the population. Broad coverage, predictable governance.
- Claude Code: targeted for senior engineers on complex refactors and platform work where reasoning depth matters.
- Governance layer: prompt logging to SIEM, 7-yr retention parity, regional tenancy review, credential-leak detection rules.
- Context-first wiring: service catalog and ownership feed agents — agents don't infer in a vacuum.
- Measurement: AI agent correctness tracked alongside DORA. Adoption ≠ value.
§ 06 — CLOSING
What this adds up to.
Companies don't fail because they lack data. They struggle because the data lacks meaning. The platforms that win the next decade are the ones that treat context as architecture — designed in, not bolted on — so that engineers ship faster, agents reason cleanly, and the business can finally trust what it sees.
That is the platform I want to build.
Brian Uckert · May 2026