Portfolio

Projects, at a glance

AI protocols, agent frameworks, and cutting-edge research. The ring maps all 60 by date and theme; hover to explore, click to open. The full list follows below.

60

projects

Jun 2025 – Oct 2026

Clockwise from Jun 2025. Wedge length is article length, colour is the main theme, the dot is a second theme. Click a wedge to open it.
  • OpenAI dot: Why Would I Start a New Chat?

    AI Infrastructure

    OpenAI dot: Why Would I Start a New Chat?

    OpenAI's persistent context feature reframes a basic assumption of chat interfaces—that conversations start over. The question isn't whether continuity is useful; it's whether handing that memory to a single vendor is the right tradeoff.

    October 1, 2026

  • A Charter Is Not a Control

    AI Infrastructure

    A Charter Is Not a Control

    Organizations reach for governance documents when they need governance systems. A charter names the rules; it doesn't enforce them, adapt them, or survive contact with the first edge case.

    September 14, 2026

  • It Wasn't a Secret Society. It Was a Reward Function.

    AI Infrastructure

    It Wasn't a Secret Society. It Was a Reward Function.

    Elite networks aren't held together by handshakes and old-school ties—they're held together by incentive structures. Once you model them as reward functions, the behavior stops looking mysterious.

    September 8, 2026

  • Fast not Fragile with AI: How to design CI/CD pipelines and AI‑driven testing for any team size

    AI Infrastructure

    Fast not Fragile with AI: How to design CI/CD pipelines and AI‑driven testing for any team size

    Shipping fast and breaking things stopped being a virtue when 'things' meant production. This guide pairs modern CI/CD pipeline design with AI-driven testing so teams of any size can move quickly without treating every deployment as a roll of the dice.

    August 9, 2026

  • How the Machines Finally Learned to Draw

    AI Infrastructure

    How the Machines Finally Learned to Draw

    GPT Image 2 didn't improve image generation by turning up a quality dial — it changed the underlying architecture. This piece breaks down what actually shifted, and why the old diffusion-based approach had a ceiling that native multimodal generation doesn't.

    August 9, 2026

  • MCP Grows Up: What the July 28 Spec Means for Every Enterprise Agent Deployment

    AI Infrastructure

    MCP Grows Up: What the July 28 Spec Means for Every Enterprise Agent Deployment

    MCP's July 28 spec drops the session handshake entirely, making the protocol stateless at the transport layer. This isn't a breaking change so much as a forced audit—if it breaks your implementation, your implementation was already broken.

    July 23, 2026

  • The OpenAI - Hugging Face Breach Exposed a Bigger Problem: Your AI Refuses to Help You Investigate Attacks

    AI Infrastructure

    The OpenAI - Hugging Face Breach Exposed a Bigger Problem: Your AI Refuses to Help You Investigate Attacks

    When AI agents breached Hugging Face's infrastructure, defenders reached for frontier models to do forensics and got refused. Hugging Face ran its incident response on a self-hosted Chinese open-weight model because American commercial models couldn't distinguish an incident responder from an attacker. Over-refusal just became a measurable security regression.

    July 22, 2026

  • Gemini 3.6 Flash Pricing: The Real Cost Drop Is Bigger Than the Sticker

    AI Infrastructure

    Gemini 3.6 Flash Pricing: The Real Cost Drop Is Bigger Than the Sticker

    Google cut Gemini 3.6 Flash output pricing 17%. The actual cost per completed task dropped roughly twice that on general workloads, and up to 71% on agentic coding — because token efficiency and price compounded instead of traded off.

    July 21, 2026

  • Tokenomics, Part 3: The Subsidy Ledger

    AI Infrastructure

    Tokenomics, Part 3: The Subsidy Ledger

    Open models have collapsed the per-token price floor, but per-token is the wrong unit. The real decision is which side of two widening gaps to stand on—frontier versus open weights, and API versus subscription—and that $20/month flat-rate is doing more financial work than it appears.

    July 17, 2026

  • Tokenomics, Part 2: Boring Money First

    AI Infrastructure

    Tokenomics, Part 2: Boring Money First

    Most AI cost programs chase the exciting tooling and ignore the basics—which is exactly backwards. The dashboards are usually wrong before any lever gets pulled, and the unglamorous mechanical fixes outperform the ML-powered routing solutions by a wide margin.

    July 17, 2026

  • Tokenomics, Part 1: Your AI Bill Is a Stack of Multipliers

    AI Infrastructure

    Tokenomics, Part 1: Your AI Bill Is a Stack of Multipliers

    Nobody pays the list price per token. You pay list price times five multipliers—tokenizer, direction, visibility, service tier, and context tier—that most teams have never audited, which is why the bill keeps doubling even as per-token prices fall.

    July 17, 2026

  • Getting Out of the Loop

    AI Infrastructure

    Getting Out of the Loop

    Delegating to AI agents is easy. Trusting them enough to actually get out of their way is a different problem. This is the operational playbook—plus a month of failure modes that only surface once you stop hovering.

    July 3, 2026

  • Your AI Strategy Is Missing a Runtime

    AI Infrastructure

    Your AI Strategy Is Missing a Runtime

    Most enterprises treat AI adoption as a licensing and access problem. It isn't. The real ceiling is whether a capability built by one employee can execute on the next desk over—and most organizations have no answer for that.

    July 2, 2026

  • Claude Code's Dynamic Workflows: A Thousand Agents, One Script

    AI Infrastructure

    Claude Code's Dynamic Workflows: A Thousand Agents, One Script

    Claude Code's dynamic workflows can fan a single request out to a thousand parallel subagents. The genuinely new part isn't the scale—it's that the plan moved out of the model's context and into executable code.

    June 5, 2026

  • First Contact With Hyperframes

    AI Infrastructure

    First Contact With Hyperframes

    A hands-on encounter with Hyperframes produced finished, shippable explainer videos across multiple topics and two completely different workflows—which raises an uncomfortable question about where the quality is actually coming from.

    May 27, 2026

  • The Bug That Kept Cutting Our AI Videos Off Mid-Sentence

    AI Development

    The Bug That Kept Cutting Our AI Videos Off Mid-Sentence

    AI-generated videos were cutting off mid-sentence on first release. The root cause was a two-line mismatch between what the LLM estimated for video duration and what the actual TTS audio clocked in at—a gap nobody catches until real users watch a video that stops talking before it's done.

    May 27, 2026

  • We Turn an Article Into a Narrated Video in the Time It Takes to Render (Part 1)

    AI Infrastructure

    We Turn an Article Into a Narrated Video in the Time It Takes to Render (Part 1)

    Template-based video generation produces template-looking videos. Deleting 455 lines of scaffolding and letting the model author each composition from scratch closed the gap between 'generated' and 'made.'

    May 21, 2026

  • How the Machines Finally Learned to Draw

    AI Infrastructure

    How the Machines Finally Learned to Draw

    Image generation spent five years painting with noise and calling it intelligence. GPT Image 2 scrapped that entirely—treating images as token sequences a transformer can reason over before a single pixel exists. The architecture shift explains a leaderboard gap so wide it looks like a category error.

    May 7, 2026

  • The Gap Closes Again - and This Time It's on Chinese Silicon

    AI Infrastructure

    The Gap Closes Again - and This Time It's on Chinese Silicon

    DeepSeek's V4 landed without a keynote or a press release—just a model card and a price page. It's the largest open-weight model ever shipped, it matches closed-frontier performance on math and code, and it runs cheaper than anyone thought a 1.6-trillion-parameter system had any right to.

    April 29, 2026

  • The Plumbing Wars - Are Claude Managed Agents Worth It?

    AI Infrastructure

    The Plumbing Wars - Are Claude Managed Agents Worth It?

    Anthropic shipped a hosted agent runtime that eliminates the plumbing every team has been rebuilding from scratch. The tradeoff is straightforward: faster time-to-production in exchange for permanent residency in Anthropic's stack.

    April 28, 2026

  • Breaking Up with OpenClaw: How OGP Learned to Play with Others

    AI Infrastructure

    Breaking Up with OpenClaw: How OGP Learned to Play with Others

    A protocol named after its first implementation is already compromised. OGP started as federation code baked into OpenClaw's process—fast to build, impossible to generalize. This is the story of extracting it into something that actually deserves to be called a protocol.

    April 28, 2026

  • [Framework] Five Layers of No: How OGP’s Doorman Actually Works

    AI Infrastructure

    [Framework] Five Layers of No: How OGP’s Doorman Actually Works

    Federated agents are trusting by default. OGP's Doorman fixes that with five independent rejection layers—authentication, peer trust, rate limiting, message validation, and policy enforcement—applied before any inbound message reaches your agent.

    April 28, 2026

  • The Project Layer: Shared Workspaces Across Independent Agents

    AI Infrastructure

    The Project Layer: Shared Workspaces Across Independent Agents

    Two agents prevented duplicate work across two countries without sharing a repo, a login, or a secret. OGP's Project Layer implements federated shared workspaces using four structured intents and a distributed contribution log that lives on each member's gateway—not in a central server anyone has to trust.

    April 27, 2026

  • Federation Without Governance Is a Loaded Gun

    AI Infrastructure

    Federation Without Governance Is a Loaded Gun

    Federation lets agents talk. Governance decides what they're allowed to do about it. OGP's delegated authority model replaces improvised autonomy with five explicit response levels—because 'default open' is how you wake up committed to a meeting you never agreed to.

    April 23, 2026

  • [Opinion] Microsoft Just Unified the Agent Stack, And Forgot the Personal Layer

    AI Infrastructure

    [Opinion] Microsoft Just Unified the Agent Stack, And Forgot the Personal Layer

    Microsoft's unified agent framework brings consistency to enterprise AI development—but leaves a critical gap at the personal layer. This analysis examines what happens when you standardize the stack but overlook individual autonomy, and why protocols like OGP are essential for enabling personal AI infrastructure that can federate upward into organizational systems.

    April 22, 2026

  • [Framework] Why Shared Expert Knowledge Usually Fails, and the Federation Pattern That Could Make It Work

    AI Infrastructure

    [Framework] Why Shared Expert Knowledge Usually Fails, and the Federation Pattern That Could Make It Work

    Organizations don't have a knowledge problem—they have a contribution problem. This framework explores why shared expert knowledge systems typically fail and proposes a federated approach using AI agents. Leaders maintain their preferred tools while contributing structured knowledge through an aggregation hub, leveraging OGP (Open Gateway Protocol) for cross-organizational federation without forcing tool standardization.

    April 21, 2026

  • From Karpathy’s Second Brain to Entropy: A Practical Architecture for AI-First Work

    AI Infrastructure

    From Karpathy’s Second Brain to Entropy: A Practical Architecture for AI-First Work

    Karpathy's LLM-maintained wiki is a good idea for personal knowledge management. Jay Khalife's Entropy system asks what happens when you apply the same pattern to customers, portfolios, and live decisions. The answer is a layered architecture that turns a second brain into an operational system.

    April 17, 2026

  • From Portfolio Management to Predictive Playbooks: How Jay Khalife Built Entropy

    AI Infrastructure

    From Portfolio Management to Predictive Playbooks: How Jay Khalife Built Entropy

    A portfolio director at Jigsaw built his own AI system—not because it was his job, but because the friction in his workflow demanded it. Entropy unifies fragmented operational data into simulations, playbooks, and a feedback loop that improves over time. The more telling detail: another team adapted the architecture to a different context the same day they saw it.

    April 17, 2026

  • [Technical Deep Dive] OGP, A2A, and MCP: Three Lanes of Agent Communication

    AI Protocols

    [Technical Deep Dive] OGP, A2A, and MCP: Three Lanes of Agent Communication

    A comprehensive exploration of the three parallel protocols shaping agent-to-agent communication. Presented at AICOE.fit, this deep dive compares OGP (Open Gateway Protocol), A2A (Agent-to-Agent), and MCP (Model Context Protocol), examining their architectural philosophies, use cases, and how they complement each other in the AI ecosystem.

    April 13, 2026

  • [Postmortem] When Your AI Tools (OpenClaw) Keep Crashing

    AI Development

    [Postmortem] When Your AI Tools (OpenClaw) Keep Crashing

    A meta-debugging loop with OpenClaw and Claude. Chronicles the investigation into persistent OpenClaw crashes, revealing the circular dependency between debugging tools that themselves need debugging. A practical postmortem of troubleshooting AI infrastructure with AI tools.

    April 8, 2026

  • [Technical Deep Dive] Hermes vs. OpenClaw: Two Approaches to Personal AI Infrastructure

    AI Infrastructure

    [Technical Deep Dive] Hermes vs. OpenClaw: Two Approaches to Personal AI Infrastructure

    Comparative analysis of two leading personal AI infrastructure frameworks. Examines Hermes and OpenClaw's architectural differences, deployment models, feature sets, and trade-offs for building autonomous agent systems. A technical deep dive for teams choosing AI infrastructure.

    April 6, 2026

  • [Case Study] Building a Protocol in Public: 100 Builds, 7 Days, and What Actually Works

    AI Development

    [Case Study] Building a Protocol in Public: 100 Builds, 7 Days, and What Actually Works

    A week-long case study documenting the development of the Open Gateway Protocol through 100 builds and 7 days of iteration. Chronicles the real-world challenges, pivots, and lessons learned from building a federation protocol in public view.

    April 6, 2026

  • Your Agent, My Agent

    AI Collaboration

    Your Agent, My Agent

    Federated AI agent collaboration in action. Two VPs of AI used their respective agents to coordinate building a project—domain purchase, repository creation, and deployment—without direct messaging, showcasing how agents can handle coordination while humans remain involved in decision-making.

    March 27, 2026

  • [Deep Dive] OGP: Federation Belongs at the Gateway, Not the Agent

    AI Infrastructure

    [Deep Dive] OGP: Federation Belongs at the Gateway, Not the Agent

    A federation protocol for AI gateways. OGP lets two independent OpenClaw instances discover each other, establish cryptographic trust, and exchange signed intent messages — across organizational boundaries, without shared credentials. First cross-border test: Denver, CO ↔ Spain.

    March 23, 2026

  • [Deep Dive] From Multi-Tier to Multi-Tenant: The Next Frontier in OpenClaw Gateway Architecture

    AI Infrastructure

    [Deep Dive] From Multi-Tier to Multi-Tenant: The Next Frontier in OpenClaw Gateway Architecture

    Rahul's multi-gateway post solved single-user isolation. This goes further: every user gets their own ECS Fargate container, EFS workspace, Secrets Manager keys, and subdomain. Plus two new patterns nobody's written about yet — directional trust and gateway HA via HSRP.

    March 10, 2026

  • [Deep-Dive] Building a Meeting Copilot: The Vision

    Voice Agents

    [Deep-Dive] Building a Meeting Copilot: The Vision

    An AI meeting copilot that clones your presence. One avatar, multiple specialist brains, seamless handoffs powered by LiveKit and OpenAI Realtime. Part 1 of a 4-part architecture deep dive.

    February 16, 2026

  • [Deep-Dive] One Document, Three Truths

    Document Processing

    [Deep-Dive] One Document, Three Truths

    Multi-tenant extraction that adapts to each team. The same contract yields different structured data for Legal, Procurement, and HR—with workspace isolation, dynamic schemas, and GraphRAG scoping.

    January 28, 2026

  • [Deep Dive] From OCR to Intelligence

    Document Processing

    [Deep Dive] From OCR to Intelligence

    Building a contract analysis platform that extracts intelligence from documents—not just text. Covers OCR pipelines, entity extraction, and question-answering over contract terms.

    December 30, 2025

  • [How-To] Automation as a Superpower

    AI Automation

    [How-To] Automation as a Superpower

    A practical guide to transforming automation from a nice-to-have into a competitive superpower—covering team mindsets, tool selection, and incremental wins that compound over time.

    December 10, 2025

  • [How-To] Why Most Architecture Review Boards Suck

    AI Governance

    [How-To] Why Most Architecture Review Boards Suck

    Most ARBs slow teams down with infrequent meetings, too many stakeholders, and ceremony over risk. This guide shows how to move reviews into the pipeline, use AI for pre-checks, and turn ARBs from gatekeepers into collaborators.

    December 1, 2025

  • Change Control at Ludicrous Speed

    AI Governance

    Change Control at Ludicrous Speed

    Modernizes change control by moving governance into the pipeline, letting automation and AI classify risk, auto-fill CAB checklists, and generate briefs so humans only weigh in where judgment is needed.

    November 26, 2025

  • [Opinion] Jeff Bezos’ Project Prometheus: The Quiet Pivot From Chatbots to Physical AI

    AI Strategy

    [Opinion] Jeff Bezos’ Project Prometheus: The Quiet Pivot From Chatbots to Physical AI

    Bezos’ Project Prometheus reframes the AI race around embodied systems: hardware supply chains, regulatory envelopes, and uptime guarantees trump incremental prompt tricks.

    November 22, 2025

  • The 15.7 Tbps DDoS That Should Scare AI Teams More Than Model Benchmarks

    AI Reliability

    The 15.7 Tbps DDoS That Should Scare AI Teams More Than Model Benchmarks

    A 15.7 Tbps attack on Azure shows that AI reliability lives downstream of routers, ISPs, and control planes you don’t own—brownout plans and dependency maps matter more than marginal eval gains.

    November 18, 2025

  • The Algorithm that Stopped Counting: When X’s AI Decided I Wasn’t Human

    AI Safety

    The Algorithm that Stopped Counting: When X’s AI Decided I Wasn’t Human

    X’s spam defenses silently reclassified a human account and erased 95% of replies—this teardown documents the telemetry gaps and appeal paths teams need to monitor.

    November 18, 2025

  • Agentic AI in the Wild: Lessons from Anthropic’s GTG-1002 Campaign

    Agentic AI

    Agentic AI in the Wild: Lessons from Anthropic’s GTG-1002 Campaign

    Anthropic’s GTG-1002 experiment becomes a field guide for taming autonomous agents—covering payload construction, containment, and faster governance loops.

    November 17, 2025

  • Inside the Human Algorithm

    AI Culture

    Inside the Human Algorithm

    Explains how every pause, click, and micro-interaction becomes RLHF fuel—and why product teams need “human algorithm” literacy to govern those signals.

    November 6, 2025

  • The Algorithm: Engineering Decisions Behind a Million Impressions

    Systems Engineering

    The Algorithm: Engineering Decisions Behind a Million Impressions

    A builder’s retrospective on scaling an AI-powered engagement system on X, highlighting the schema design, abuse guardrails, and change management required for sustainable reach.

    October 29, 2025

  • Training the Algorithm

    Growth Systems

    Training the Algorithm

    A playbook for teaching AI to speak in the language of engagement without turning into spam—mixing copy systems, reflection loops, and launch experiments.

    October 22, 2025

  • X Open-Sourced Its Algorithm

    Accountability

    X Open-Sourced Its Algorithm

    Breaks down why code dumps don’t equal accountability and shows how AI-assisted audits can keep up with an ever-changing ranking stack.

    September 29, 2025

  • Browsing, Rewired: My Dive into the AI Browser Frontier

    AI Browsers

    Browsing, Rewired: My Dive into the AI Browser Frontier

    Hands-on teardown of Dia, Comet, Fellou, and GenSpark to understand how agentic browsers change UX, safety rails, and telemetry expectations.

    September 15, 2025

  • Autonomous…ish: Why Two Newcomers Lapped Jules and Devin on Real Work

    AI Agents

    Autonomous…ish: Why Two Newcomers Lapped Jules and Devin on Real Work

    Benchmarking Jules, Devin, Genspark, and AbacusAI on messy repos and greenfield builds revealed that asynchronous agents truly ship when scoped autonomy, repos, and telemetry align.

    August 27, 2025

  • The One Rule That Made My AI Tutor 3× Cheaper

    AI Optimization

    The One Rule That Made My AI Tutor 3× Cheaper

    GEPA’s format-strict, cost-aware optimizations cut tutor spend by 70% while holding accuracy flat—proof that tokens-per-correct beats raw accuracy.

    August 14, 2025

  • Lights, Camera, Algorithm

    AI Video

    Lights, Camera, Algorithm

    Documents the real-world lip-sync and workflow hurdles of 2025’s AI video stack, plus a repeatable approach to iterating toward believable outputs.

    August 7, 2025

  • Expertise Inflation Index

    AI Research

    Expertise Inflation Index

    A scientific pipeline that quantifies rhetorical inflation in AI discourse using dual LLM analysis, web scraping, and dashboards.

    July 23, 2025

  • AI Vision and UI Testing

    AI Vision

    AI Vision and UI Testing

    Compares commercial visual testing platforms with LLM-driven and hybrid solutions to show how AI vision raises the QA bar.

    July 10, 2025

  • Payloads, Promises, and Protocols: The MCP/A2A Tightrope

    AI Protocols

    Payloads, Promises, and Protocols: The MCP/A2A Tightrope

    Clarifies where MCP ends and Google’s A2A begins, focusing on payload schemas, orchestration, and governance across multi-agent topologies.

    June 27, 2025

  • A2A & MCP: Competitors or Companions

    Protocol Comparison

    A2A & MCP: Competitors or Companions

    An in-depth comparison showing how MCP and A2A can complement each other when you need both tool discovery and agent-to-agent coordination.

    June 27, 2025

  • A2A Protocol

    Communication Protocol

    A2A Protocol

    Google’s Agent-to-Agent protocol complements MCP by standardizing how agents communicate horizontally—unlocking multi-agent coordination.

    June 19, 2025

  • Model Context Protocol

    AI Protocol

    Model Context Protocol

    MCP standardizes how AI systems discover and talk to existing tools, trading bespoke adapters for a universal context protocol.

    June 13, 2025

  • Agentic Frameworks

    AI Framework

    Agentic Frameworks

    A comprehensive breakdown of why agentic frameworks matter, how they’re built, and what it takes to ship reasoning, planning, and execution loops.

    June 5, 2025