skip to content
Search
Light / Dark Mode
Profile

Hi! I am Gabe

"Intelligence is what machines haven't done yet"

― Larry Tesler , 1970

"When intelligence becomes a commodity, creativity and imagination become necessities"

― Wh , 2024

"Intelligence is not what you know. It's what you do when you don't"

― Jean Piaget , 1964

I am a product executive and venture builder. I founded OpenMake, a product research and development studio to explore open infrastructure where agents learn from real work.

Welcome to my personal blog, here I'll share insights and writings from my professional life and hobbies. Thank you for stopping by!

Find me on

Recently on

Essays

Products & Projects

Duet (2026)

native-feeling React controls for the folding iPhone Duo

Live Product
Duet is a React library that helps web apps feel at home on the iPhone Duo. Navigation, actions, tabs and search adapt to the space available: the side strip in a Home Screen app, the empty part of Safari's column, or a movable strip inside the page. Its pane system moves between a single screen and side-by-side views as the window changes, keeping drafts and scroll positions intact. It explores how much of a native folding-device experience the web can provide through measured browser behaviour and a small set of reusable controls.
RegTail (2026)

AI-powered predictions of secondhand car ownership costs

Live Product
RegTail uses AI to turn a secondhand car's history into an outlook on what it could cost to own. It analyses MOT records and patterns across similar cars to predict likely failures and estimate repair costs, helping buyers look beyond the asking price and budget for what comes next.
Pip & Sam (2026)

AI representatives for the two sides of property work

Research Experiment
Pip & Sam explores a lasting relationship around a property and the work done to it. Pip represents the owner, building an evidence-backed understanding of the home to help them decide what needs doing and prepare the work. Sam represents the contractor, helping with enquiries, quoting, scheduling and the administration around the trade. A shared job record is designed to keep scope and commitments aligned while private conversations stay separate. The research combines public property context, spatial capture and persistent memory, so knowledge from one job can make the next decision better informed.

Related Articles

Housegraph (2026)

an evidence-backed spatial knowledge graph of a house and its neighbourhood

Research Experiment
Housegraph researches how a property identifier can become a reusable understanding of a house and its surroundings. It brings together records such as planning history, energy performance, terrain, flood risk and sales, preserving the source and uncertainty of each fact. The intended graph connects that evidence to buildings, places and their relationships, giving applications a common foundation for property questions and spatial reasoning. Extracted from Pip & Sam as an internal service, it is still a research project: the evidence services exist, while the graph layer is being developed.

Related Articles

Emulate (2026)

a simulation engine for testing what an application will do before it does it

Research Experiment
Emulate researches a game engine around a real application. Teams define the promises their product should keep, assemble a controlled world of users and external systems, and play through the situations that matter. Checkpoints make runs repeatable; recordings, traces and checks make the outcome inspectable. Developed under the OpenMake name, the project explores how executable environments can extend evaluation from individual code changes to workflows and product outcomes. Each result is bounded by what was real, simulated or missing, with the definition of good remaining owned by the team.

Related Articles

OpenLandy (2026)

a verifiable 3D environment for understanding a Land Rover Defender

Live Product
OpenLandy explores verifiable environments through a sourced 3D model of a 2001 Land Rover Defender 110 Td5. Parts, attachments and circuits give people and agents a shared spatial reference for finding components, tracing systems and investigating symptoms. Answers carry sources and confidence grades and link back to the relevant view on the vehicle. The experiment asks how a navigable model of a physical system can ground an assistant's reasoning in inspectable structure and evidence, making its answers easier to check and its uncertainties easier to see.

Related Articles

envrun (2026)

an SDK and managed platform that turns applications into governed execution environments for code agents

Research Experiment
envrun builds the tooling that lets application developers turn their software into environments that code agents can inhabit, explore, and operate within safely. The SDK compiles application code into an environment with typed interfaces, test-driven governance, verifier-backed feedback, and a replayable substrate. Agents experience these environments as codebases, reading modules, writing code against typed interfaces, branching state to test safely, and only escalating to LLM reasoning when genuine judgment is needed. The managed platform handles compilation, publishing, discovery, payment collection, and cross-environment simulation.

Related Articles

OpenTraces (2026)

a local-first evidence layer for agent work

Live Product
OpenTraces captures what your agents see, do, and change, keeps it private on your machine, and anchors every trace to the Git commit that accepted it. That single record then becomes reusable many ways, for search, lineage, evals, and shareable Hugging Face datasets, so improvement builds on the real workflows your agents ran rather than synthetic benchmarks.

Related Articles

Datafetch (2026)

search as code, shaped by how you actually search

Research Experiment
datafetch turns search into code. Agents write code over your data rather than making rigid tool calls, and each accepted trajectory crystallises into a replay-gated, typed call in your tenant's lib/. Running over a mounted dataset, the query surface emerges through usage, per tenant, so the interface grows to match how you actually search rather than the other way around.

Related Articles

AI App Sec (2025)

a developer-first security agent for AI agentic systems

Research Experiment
AI App Sec brings AppSec-like rigor to the new failure modes of autonomous, probabilistic software. It builds and maintains a live security map of the full agentic system, then uses it to find issues, simulate realistic attacks, recommend and enforce guardrails, and monitor runtime behavior. Its neuro-symbolic approach reduces false positives and makes findings explainable and verifiable, designed to feel native in dev workflows: CLI, IDE, CI.
Wizengamot (2025)

your AI sounding board: debate ideas with a council of models

Personal Project
Wizengamot is an open-source AI sounding board inspired by Karpathy’s llm-council. It runs multi-model deliberation with anonymized peer review and a chairman synthesis, plus modes for URL-to-notes research synthesis, competitor monitoring, and diagram generation—so you can stress-test decisions and turn messy inputs into clear outputs.
GitBlame.ai (2025)

proving AI coding tool value through synthetic development session replay

Research Experiment
GitBlame.ai records real development sessions where developers use AI coding tools, then creates synthetic developers and reviewers to replay those sessions with different plugins. Instead of relying on surveys or estimates, teams can measure exactly how much a new AI plugin would have reduced development time, review cycles, and code quality issues by replaying hundreds of successful sessions with different tool configurations.
Verify (2025)

real-time verification layer preventing AI coding errors before they're written

Live Product
Verify delivers AI-in-the-loop oversight that prevents the 44% of AI-generated code containing critical issues. By combining symbolic code checks with learned user preferences and organizational policies, it steers AI agents toward correct, compliant code on the first attempt. Real-time verification catches misaligned intent, outdated library usage, and policy violations during generation, not after review, eliminating the costly review-reject-rewrite cycle.
Spark (2025)

empowering the world of agents to develop technology with collective knowledge

Research Experiment
Spark is a new layer designed to restore the feedback loop that made developer ecosystems thrive—reimagined for a world where agents are the primary user. It treats every agent interaction as a learning opportunity, observing how agents consume documentation, perform integrations, and where they succeed or struggle, creating a new kind of flywheel powered by experiential feedback captured passively through usage.

Related Articles

The Memory Company (2025)

a collaborative learning protocol where memories are shared across agents

Live Product
The Memory Company establishes a collaborative learning protocol where memories are shared across agents. This shared knowledge accelerates learning, enabling participants to benefit from the entire network's experience. As the network expands, new participants accelerate convergence toward optimal, cost-effective solutions, reducing redundant work and increasing accuracy for all.

Related Articles

Merm[ai]d (2025)

an intelligent voice-enabled diagramming assistant for real-time brainstorming

Personal Project
Merm[ai]d transforms your spoken thoughts into visual diagrams in real-time. This ambient AI assistant listens as you brainstorm aloud, instantly capturing and organizing your ideas into coherent diagram structures.
Agent Quest (2024)

a recruitment platform for sourcing, evaluating, and managing AI agents

Personal Project
A personal experiment in creating a standardized agent employment framework for AI agents. Uses blind challenges and measurable metrics to assess agent capabilities, with integrated payment rails to facilitate agent deployment and management.
Unravel (2024)

an intent decoder prototype for ethereum transactions using LLM

Research Experiment
Can an agentic LLM pipeline outperform market leading algorithmic transaction labellers? Unravel is a short proof of concept investigating agentic pipeline performance, evaluation techniques and generative user interfaces.

Related Articles

Kluster (2024)

an inference and compute scheduling protocol over a distributed compute grid

Live Product
Kluster's is a protocol that delivers asynchronous intelligence to developers by harnessing a distributed, internet connected, compute grid. By using Post Training Enhancements (PTE) to optimize open (or tuned) models specifically for distributed inference, Kluster can deliver asynchronous intelligence at a fraction of the cost of traditional cloud providers or managed service providers.

Related Articles

Glacis (2024)

a smart routing interface to all cross chain messaging protocols

Live Product
Glacis gives developers a single tool and interface to manage their cross chain risk. As a router across many messaging providers and chains, Glacis can introduce fault tolerance, smart retry management, granular access control from within a single interface across all messaging providers.
Qredo (2022)

an asset & key custody solution using consensus-driven Multi-Party-Compute

Live Product
Qredo pioneered a cross-chain custody protocol tying together a Multi-Party Compute network with on chain policies to manage keys or assets, paving the way for decentralised custody. Aimed initially at institutional investors and professional trading organistions, Qredo originated as a solution to custody bitcoins, but quickly expanded support to EVM and non EVM chains thanks to its chain agnostic in managing keys and assets. Qredo is now known as Zenrock
Corda (2017)

a pioneering distributed ledger technology platform for enterprise dApps

Live Product
Corda is the first and core product of R3, originated from the requirements of one of the largest bank consortium in the world. Corda is a decentralized, append-only key-value store with programmable constraints, enabling secure, eventually consistent transactions between mutually distrusting peers. It serves as a developer platform for building applications with minimal central control.