Noah Santoni

Today I’m launching A2A Net, the platform for B2B SaaS to automatically optimise an AI agent from their APIs that their customers can use in their website, app, Slack, Teams, Copilot, and Gemini Enterprise! Despite “SaaSpocalypse” being one of the best terms to come out of 2026, I’m optimistic about the future of SaaS. There’s no doubt that it’s easier to build software now, but that doesn’t change economics: one company can create software for one hundred companies cheaper than one hundred companies can create that software for themselves. That one company will also be specialised, and as a result will be able to solve the problem better, provide better support, etc. What’s more, SaaS stands to benefit more than any other industry from AI, because SaaS and AI both run on computers, AI can add value to every corner of SaaS! This is especially true for SaaS categories that are used frequently or have lots of data, records, or workflows, e.g. GTM, ATS, customer success, CRM, project and work management, marketing operations, finance and accounting, customer support and service desk, HRIS, procurement, etc. That’s not to say that it’s easy to add AI to SaaS... Everyone who’s built and deployed an AI agent to production knows: it’s hard to build and maintain production-ready agents. Prompt changes fix one thing and break another, and building good evals is genuinely difficult. Coding agents help, but require constant input from an AI Engineer. Prompt optimisation algorithms like GEPA exist, but the optimiser requires a numerical score and natural language feedback. For that, you need to build good evals, which are not only difficult and time-consuming to create, but typically represent a small subset of real-world tasks. This is why constant input from an AI Engineer is required, to give feedback to the coding agent. The coding agent is the optimiser, and the AI Engineer is the evaluator. A2A Net solves this problem by taking the AI Engineer out of the loop. Agents are built and tested in the same way. But, instead of telling a coding agent what to improve, you rate the agent’s sessions. Those ratings are used to teach an LLM-as-a-judge to rate sessions like you. That is then used to automatically optimise the agent. If you have an OpenAPI spec or an MCP server, you can create an agent for your B2B SaaS in 5 minutes by going to A2A Net’s landing page, entering your website, and clicking “Create agent”! One customer found that the agent that was built for them from A2A Net’s landing page was better than the agent they had been building for months. Another deployed an agent to their customers in a week, who were using it to complete complex multi-step workflows, like uploading unstructured spreadsheets with 50K+ cells into their platform! If you’d like to chat about AI agents for B2B SaaS, please feel free to send me a message or book a meeting!

Sometimes I leave out context and just sit and watch the agents struggle. Reminds them who's still in charge.

Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.

My friend Milla Jovovich and I spent months creating an AI memory system with Claude. It just posted a perfect score on the standard benchmark - beating every product in the space, free or paid. It's called MemPalace, and it works nothing like anything else out there. Instead of sending your data to a background agent in the cloud, it mines your conversations locally and organizes them into a palace - a structured architecture with wings, halls, and rooms that mirrors how human memory actually works. Here is what that gets you: → Your AI knows who you are before you type a single word - family, projects, preferences, loaded in ~120 tokens → Palace architecture organizes memories by domain and type - not a flat list of facts, a navigable structure → Semantic search across months of conversations finds the answer in position 1 or 2 → AAAK compression fits your entire life context into 120 tokens - 30x lossless compression any LLM reads natively → Contradiction detection catches wrong names, wrong pronouns, wrong ages before you ever see them The benchmarks: 100% recall on LongMemEval — first perfect score ever recorded. 500/500 questions. Every question type at 100%. 92.9% on ConvoMem — more than 2x Mem0's score. 100% on LoCoMo — every multi-hop reasoning category, including temporal inference which stumps most systems. No API key. No cloud. No subscription. One dependency. Runs on your machine. Your memories never leave. MIT License. 100% Open Source. github.com/milla-jovovich/me…

no other way to celebrate 1 year of @unicorn_mafia than to deploy a virtual world with our pixel avatars running in parallel to an epic party @pixelsocieties