After two decades building software for enterprises and public services, I keep returning to one question: what should people be able to see, choose and control?
That question guides my work on AI: clear identities, meaningful permissions and evidence for what systems actually do.
Two explorations in trust and accountability.
Your agent should carry your trust with it. AgentHalo brings identity, memory, permissions and consent together across devices, with a record of what an agent was allowed to do—and what it did.
A convincing answer is only a beginning. Truthseek puts an agent’s claims to the test, through an evaluation harness and a companion that brings that scrutiny into the editor.
A book of real product screens.
Each page begins with the product itself. A slow, continuous turn gives every screen room to be seen.
The public products, platforms, tools and experiments in this collection.
A personal AI workspace built around user-owned models, portable knowledge and inspectable behavior.
A customer relationship product for people who want a more personal, human way to stay connected.
An India-focused civic guide that brings regional service information and official-source handoffs into one place.
Runtime FinOps for AI: metering, budgets, routing and chargeback across model providers.
A human-guided assembly space for shaping multi-agent systems from reusable parts.
A live research preview for evaluating which tasks a personal AI can safely take on. Profiles and results are synthetic fixtures, not commercial rankings.
A searchable AI-agent directory and practical privacy guides. Directory listings are marked as not reviewed.
A searchable directory of hosted MCP servers and API connectors for agents, with early access for new connectors.
An independent connector directory and workflow community for personal AI agents, with a reviewable developer submission flow.
A private AI command center for models, memory and knowledge you can inspect, move and run yourself.
A free CRM built to stay useful, open and available to everyone.
A reference platform for repeatable agent trials, evidence and release decisions, with clear limits on what it proves.
A before-and-after lab for agent-ready websites, tool contracts and evidence that separates source hints from runtime proof.
A tool for checking agent-ready websites and exploring agent workflows, with source findings kept distinct from runtime evidence.
A two-phone Safari demo that connects agents through an Inkbox A2A handoff.
Open-source foundations for composing agentic systems and workflows.
Local-first, self-hostable tools designed with privacy in mind.
An open-source agentic micro-drama production studio with evidence-gated releases.
An evaluation harness for testing the claims made by agents.
A companion for bringing claim evaluation into the editor.
Identity, memory, permissions and consent for an agent across devices.
A public experiment in verifiable proof and recognition.
A community catalog of reusable, reviewable instruction packs for focused AI-agent tasks.
A source-linked directory of AI-agent applications discovered across GitHub, the web and X.
A source-linked directory of apps built with AI coding tools, discovered across GitHub, the web and X.
A working notebook of experiments built with Google AI Studio.
An open record of prototypes, ideas and working explorations.
A small public software experiment from the wider lab.
Titles and short descriptions; details stay private.
A local-first personal AI companion for iOS and Android. The work centers on continuity, useful memory and trustworthy everyday assistance.
A civic companion for finding official services and preparing for the next step. It pairs sourced public information with private personal preparation.
A Chinese- and English-language civic guide with curated service information and official-source handoffs.
A Russian-language civic guide with practical service explanations, official links and browser-saved checklists.
A Portuguese-first civic guide for Brazil, built around practical starting points, official sources and locally saved checklists.
A guide to U.S. public services that helps people understand where to look and prepare what to ask. It complements the official agency websites.
A personal privacy and trust assistant for making sense of uncertain situations. It is designed around evidence and clear user choice.
An exploration of practical support for AI builders. The work focuses on clear budgets, inspectable outcomes and user-controlled participation.
Additional tools for estimating model costs, speed and resource needs. The public Tokenomy product is linked in the collection above.
Tools for understanding and reducing spend across coding agents. The emphasis is on diagnosis and changes the user can review.
Private infrastructure for agent identity, discovery, authorization, policy, routing and lifecycle management.
Private work on durable evidence records for agentic systems, with integrity checks and independent verification in mind.
Private revenue infrastructure for websites and AI agents, focused on clear attribution and verifiable outcomes.
Private evaluation research for personal-AI delegation, built around executable tasks and inspectable evidence.
A private, local-first assistant concept for helping people keep up with the relationships that matter to them.
A private research archive that organizes public enterprise-AI writing into practical working references.
A private exploration of an opinionated, touch-first Linux experience for phones and open hardware.
A private concept for reviewing agent identity, runtime policy and assurance evidence together. It is exploratory work, not a certification or endorsement.
Co-authored papers and contributions to shared foundations.
Discovery, capability verification and zero-trust collaboration for enterprise agents.
The infrastructure behind a globally addressable internet of agents.
An agent index built around verifiable identity and signed AgentFacts.
A foundational question about the architecture autonomous agents need.
A framework for assessing vulnerabilities in AI and agentic systems.
I’m Mahesh Lambe, a founder, engineer, researcher and angel investor in Palo Alto, California. I build software around a simple conviction: people should retain ownership of their data, their tools and the decisions made on their behalf.
For more than twenty years, I’ve worked on AI, CRM and cloud platforms for governments and global enterprises, including Microsoft, PIMCO and the NYPD. My public-sector work includes large-scale systems for more than 170 California state departments. That experience shapes my focus on reliability, accountability and freedom from vendor lock-in.
I’m the founder and CEO of Unify Dynamics, working across CRM, cloud and generative AI. I’ve founded four startups and, as an angel investor, backed more than 100 companies across AI, automation and enterprise software. Building companies and supporting other founders are both part of my work.
I’m a core contributor to MIT’s Project NANDA, exploring how agents discover one another, establish identity and collaborate across systems. I co-authored the four NANDA papers collected here, with Ramesh Raskar and collaborators, on agent naming, verified AgentFacts, adaptive resolution and Zero Trust Agentic Access.
I’m also a founding member, reviewer and contributor to the OWASP AI Vulnerability Scoring System, helping develop shared ways to assess security risks in AI and agentic systems.
My upstream proposals to WebMCP address browser-tool permissions and clearer tool-execution contracts.
My public work spans user-owned AI, CRM and model economics; civic-service guides; agent discovery, privacy and consent; developer tools; and evaluation and assurance. It includes MySovereign.ai, LoveCRM.org, Tokenomy.ai, LegoClaw.com, Plus91, ConnectorBay, Agent Privacy, Personaeval, IsWebMCP, AgentLens, AgentBump, Enterprise Agent Simulation Assurance, OpenWood, MuseTown × MuseSquare, and source-linked directories.
Alongside that work, I build private products for personal AI, regional civic services, privacy, agent control, evidence and revenue systems. The common thread is making actions inspectable, permissions revocable and ownership meaningful.
Sharing what I learn has long been part of the practice. I was among the top five global MSDN contributors for five consecutive years. Today, that continues through open-source projects, technical essays and conference talks on agent registries, AgentFacts and the architecture of an open agentic web.
Ideas from the work.
How Project NANDA could become the DNS of an agentic internet.
A technical deep dive into naming and discovery.
Giving every agent a verifiable capability record.
I gathered all 139 published articles here. Search a title or subject; each preview opens the original on LinkedIn.
139 articles
An observation on AI-assisted entrepreneurship, the homogenization of product thinking, and the dangerous disappearance of independent judgment. There is something…
Read originalYour AI agent knows a remarkable amount. But can it check? Imagine waking up on a Friday morning and asking your personal AI agent whether the air is good enough…
Read originalWhy the biggest business opportunity may not be building another AI agent, but connecting millions of them to the real world. Imagine waking up tomorrow morning to…
Read originalWhat an ordinary household repair reveals about the scale, economics and future of the agentic web. Imagine a Sunday morning when the refrigerator has stopped…
Read originalA practical tour of the isWebMCP.com platform for teams building agent-ready web experiences WebMCP adoption is easy to oversimplify. A page can have useful…
Read originalOpenAI released the token counts and skipped the invoice. Here's the arithmetic — and why the gap between compute cost and retail price is the most important number…
Read originalI started building FREE CRM around a simple frustration: too much of our relationship work is scattered across contact lists, notes, sales tools, billing systems,…
Read originalIt is Sunday morning. Somewhere, a boardroom is discussing how to own the future of artificial intelligence. Somewhere else, a student has downloaded a model,…
Read originalAn extraordinary advance in AI. A broader claim that still needs evidence. Imagine handing an AI a project that matters to you. The instructions are incomplete. Two…
Read originalYour AI agent is brilliant. Then it opens a website. Suddenly, it is back in kindergarten. It studies buttons, guesses what icons mean, wrestles with date pickers,…
Read originalMost AI procurement conversations I sit in evaluate the wrong thing. Teams benchmark model quality, price per seat, and integrations. Those are the easy questions,…
Read originalExecutive summary Every previous wave of technology locked in something you owned: your files, your rows in a database, your phone number. The AI wave locks in…
Read originalIf our interests meet,
I’d be glad to hear from you.