I deploy open-source AI agents (Hermes, OpenClaw, Paperclip) on your private infrastructure — integrated with your stack, in Hebrew and English, with accessibility built in.
We'll start with a short, clear conversation: what already exists, what's holding you back, what's urgent, and the step that delivers the most value without an unnecessary project.
Plain titles on purpose: business outcome first, technology underneath.
Audit, fixes, accessibility menu, statement, and a process aimed at high standards.
Read more →BotsBots that handle repeat questions, integrate with your stack, and leave the hard work to humans.
Read more →AI AgentsAgents that complete a task end-to-end — not just answer. MCP, tools, and real integrations.
Read more →Personal AIFocus in email, meeting summaries, less personal ops — without leaking data outside.
Read more →Agent SetupOpen-source agent platforms running on your own server — in days, not months.
Read more →Self-hostedModels, agents, and AI tools on your own servers. Ollama, vLLM, OpenWebUI — all under your control.
Read more →Understand the business goal.
What's missing, stuck, urgent.
Short, clear recommendation.
Build, fix, integrate.
Validate against real usage.
Content that answers the questions managers actually ask: what's worth it, what's risky, what to do first.
What a manager needs to know before starting an accessibility project — and the corners everyone forgets.
Questions to ask before you add a bot to your site or the company WhatsApp.
Simple checks you can run before signing up for an expensive DevOps project.
Short updates on one story worth knowing — with the practical angle, no fluff.
The interesting move isn't the tool, it's the direction: OpenAI is normalizing the idea that you wire all of your financial data into an AI assistant. Within a year, customers will expect your bot to know what they bought, when, and at what price. If you're not designing for that now, you'll be late.
Beyond the headline: it's a four-year deal, mostly model credits rather than cash. Anthropic is signaling where it sees itself in 2030 — not just an API, but infrastructure for public-sector and nonprofit work. If you sell into government, education, or healthcare, this means budget exists for pilots and a precedent already sits on the table.
Three notes: (1) Custom MCP support opens internal-tool integration without building a plugin. (2) Grok is well behind Claude and ChatGPT on enterprise connectors — this is the catch-up move. (3) Before wiring your whole Workspace in, read what Grok retains and what it doesn't — Anthropic and OpenAI are much more transparent here.
The handful I actually learn from on AI, engineering, and design. A short list — not everyone, just the ones I read and listen to.
AI tool builder · daily writer · @simonw
Near-daily notes on what's actually working with LLMs — with code, tools, and the receipts.
Evals + LLM consulting · @HamelHusain
The clearest writer on LLM evals — how to know your AI system actually works in production.
Applied ML engineer · @eugeneyan
Deep essays on shipping AI products, not just demos. The applied side of the field.
Structured outputs, RAG · @jxnlco
Creator of Instructor. Sharp, hands-on writing on RAG, agents, and AI engineering.
Latent Space · AI Engineer · @swyx
Latent Space podcast + newsletter that map the frontier of AI engineering as it ships.
Designer · anthropologist · @Mappletons
Brilliant essays and illustrations on how AI is reshaping interfaces and knowledge work.
AI educator · ex-OpenAI/Tesla · @karpathy
The clearest first-principles teacher on how LLMs work, from tokenizers to transformers.
AI educator (IL) · @aiwithamitay
Practical AI for Israeli companies — Substack notes and corporate workshops.
Machine & Deep Learning Israel · MDLI
Runs Israel's largest ML/AI community — meetups, annual reviews, and the central hub for talent.
Initial audit takes a few days. Fixes depend on site size, but for most businesses it's a few weeks — not half a year.
In some cases yes, but mostly it solves the repeat questions and frees the human for the hard ones. Worth mapping out before you build.
We start with a discovery call and a focused proposal. I don't believe in selling a project when a fix is enough — or selling a fix when the system really needs replacing.
Yes. A lot of my work is architecture alignment and prioritization with the existing team, not in place of it.
Site, accessibility, bot, IT, cloud, an existing product or a new idea — we'll find the right next step without overloading.