2026 In Review · Life & Milestones
Where Life Stands in 2026
On building platforms, carrying 130B+ tokens, leading through architecture, and staying close to the code.
2026 has been one of the most intellectually demanding and deeply rewarding years of my career. It was the year our Agentic AI Platform moved from early prototypes into the operational central nervous system for our enterprise. Systems that were architectural sketches eighteen months ago now carry billions of tokens and hundreds of thousands of tool calls every week.
As an AI Architect and company Distinguished Engineer, leadership at this scale looks different than managing teams. It means setting architectural standards that outlive reorganizations, writing clear decision records that give engineering teams autonomy, and mentoring senior engineers so they become owners of the systems they build.
Yet, I have made a deliberate commitment to never leave the terminal. In 2026 alone, I authored more than 1,200 commits across our gateways, networking infrastructure, and compliance engines. Staying close to the code is what keeps architecture honest. It tells you immediately when an abstraction leaks, when a network packet drops, or when a Kubernetes autoscaler fails to bin-pack efficiently.
Outside of production deployments, 2026 has also been a year of deep theoretical curiosity. I spent evenings and weekends breaking down the core mathematics behind generative AI — probability metrics, diffusion theory, and f-divergences — because understanding the foundational equations makes us much better system architects.
Scale
2026 by the Numbers
Tokens served through the model gateway
Tool calls brokered across 120+ MCP servers
AI agents active weekly across business lines
Compliance certifications (ISO 27001, SOC 2, ISO 42001)
Production commits shipped across 60+ repos in 2026
Years designing distributed platform architectures
Chronology
2026 Quarterly Milestones
Interactive Agent UIs, IssueOps Automation & Theoretical Foundations
Pushed the boundaries of what conversational AI can deliver by moving beyond text into interactive interfaces, automating cloud delivery, and diving into generative AI mathematics.
Beyond Text: Interactive Micro-Frontends with MCP Apps
Chat alone is too narrow for complex enterprise workflows. Designed and deployed MCP Apps — interactive UI widgets (tree inspectors, charts, usage meters, and approval forms) running inside sandboxed client frames with bi-directional state synchronization.
IssueOps and GitOps Self-Service Platform
Eliminated infrastructure ticketing bottlenecks by creating an automated onboarding pipeline. Developers submit structured GitHub issues to provision container registries, KMS keys, OIDC-federated IAM roles, and ArgoCD application manifests with automated team approvals.
Mathematical Foundations of Generative AI
Dedicated deep focus to the theoretical underpinnings of modern AI: researching and writing about f-divergences, diffusion dynamics, and probability distance metrics to understand why models behave the way they do.
Spot Compute at Scale, Unified Telemetry & Triple Compliance as Code
Scaled the platform to handle massive background AI workloads while establishing automated compliance as code across ISO 27001, SOC 2, and the new ISO 42001 AI standard.
10,000+ AI Batch Jobs on EKS Auto Mode & Karpenter
Built the high-throughput batch compute architecture for document parsing, chunking, and embedding. Combined Amazon EKS on EC2 Auto Mode with Karpenter Spot provisioning, cutting batch compute expenses by 70–90% while using SQS and KEDA for scale-to-zero autoscaling.
Measure Once, Distribute Everywhere: Central GenAI Telemetry
Architected an OpenTelemetry pipeline that ingests verbose multi-step agent traces once and forks them into real-time operational APM, an immutable token ledger for chargeback, and a ClickHouse columnar OLAP data store handling 500M+ rows per day.
Continuous Compliance as Code (ISO 27001 & SOC 2 Type II)
Replaced manual screenshot collection with automated GitHub posture auditing and idempotent ruleset reconcilers across 60+ repositories. Verified branch rulesets, non-trivial CODEOWNERS, and team approval gates continuously with versioned JSON audit evidence.
Pioneering ISO/IEC 42001 (Artificial Intelligence Management System)
Established the enterprise governance architecture for the first international AI management standard. Formalized AI System Impact Assessments (AIIA), AI Nutrition Labels, agent reasoning trace auditability, and Third-Party Risk Management for foundation model providers.
Architecting Platform Engineering
Established the protocol standard for agent tool connectivity and scaled platform microservices with robust egress protection.
High-Density Kubernetes Networking Architecture
Architected high-density pod networking and custom subnet routing to support thousands of concurrent batch ingestion pods with zero overlay encapsulation penalty.
Enterprise Model Context Protocol (MCP) Tool Broker
Created an enterprise MCP gateway that dynamically parses OpenAPI specifications from internal services and exposes them as typed, discoverable agent tools. Implemented recursive JSON schema dereferencing ($ref) and fine-grained access control lists.
Egress Filtering and Blast Radius Isolation
Segmented platform microservices into dedicated virtual networks (Agent, Model, Knowledge, Tools). Deployed network firewalls and waypoint proxies to inspect outbound internet egress, preventing Server-Side Request Forgery (SSRF) during agent tool executions.
Multi-Cloud Model Gateway V2, Token Vault & Cloud Consolidation
Set the platform foundation by building a unified model gateway, solving delegated identity, and executing a strategic cloud consolidation.
Multi-Cloud Model Gateway Architecture
Engineered an OpenAI-compatible model gateway abstracting 60+ foundation models across Azure OpenAI, Google Cloud Vertex AI, Anthropic, and open weights. Added cross-cloud failover, circuit breakers, upstream Retry-After forwarding, and regional TTFT/TPOT latency telemetry.
The Token Vault: Safe OAuth2 Delegation for AI Agents
Solved the security risk of user-delegated tool calls. Built an isolated Token Vault service that securely handles OAuth refresh cycles, session cookies, and scope isolation, allowing agents to act on behalf of users in tools like Google Workspace and GitHub without credential leakage.
Strategic Cloud Infrastructure Consolidation
Authored core Architecture Decision Records (ADRs) consolidating primary platform compute, storage, and databases on AWS to establish unified operational standards, while maintaining a clean multi-cloud abstraction for foundation model APIs.
Reflections
What 2026 Taught Me
Platforms before features
When every application team tries to build their own model integrations, authentication, and tool routing, you end up with fragmented security and unmanaged cloud spend. Solving the hard infrastructure problems once creates a shared foundation where dozens of teams can innovate safely.
Measure before you scale
Generative AI is inherently non-deterministic. Without objective quality benchmarking, token cost ledgers, and latency observability, you cannot tell if an update helped or harmed your users. Measurement turns AI quality from a subjective debate into an engineering metric.
Make the safe path the easy path
Compliance, security, and responsible AI guardrails fail when they feel like bureaucratic friction. When you automate compliance into GitHub rulesets, merge checks, and pre-configured cloud templates, teams get enterprise-grade security for free on day one.
Foresee and build for the quiet future
A platform architect’s greatest asset is the ability to foresee a distant future and build for it early. It is also a quiet paradox: when you anticipate a problem and engineer the solution years before the crisis hits, no one notices the disaster that never happened. But that quiet foresight is what protects an enterprise when sudden scale arrives.
Stay close to the code
Architectural authority is earned in production, not drawn on whiteboards. Even as an AI Architect setting multi-year strategy, staying hands-on in the terminal and shipping code keeps design decisions practical, honest, and grounded in real system constraints.
Ready to dive deeper?
Explore the technical deep dives explaining how we built these systems, or get in touch to discuss enterprise AI architecture.