If you want to be discovered by LLMs, you have to build data they can rely on . I think we are getting closer to have "agentic web" which is not based on scraping your website but built on top of a standard (like MCP).

AI startup CFO, CEO, CMO, and CTO

every evals/analytics startup is going through a onetime generational upgrade into a continual learning platform in 2026 many will fail but as always the tasteful ones win

One of the best things students and colleges can do is not bail on learning and teaching the fundamentals of any given domain. AI will trick you into thinking you don’t need to go deep in a particular area, but that’s wrong. The expert with AI is always going to be far more capable than the novice. Those that can steer AI agents properly, figure out how to evaluate their work, fix their mistakes, and incorporate their work into a workflow will always be the most potent users of these tools. The experienced software developer that’s built and scaled complex systems using agents outrun someone just vibe coding. The designer that uses AI will build far better products and campaigns than anyone else. The banker or analyst that understands financial models will be able to pull off far more with agents. Despite some of the rhetoric in the valley that this is less implement now, that couldn’t be further from the case. Don’t give up on going deep in your craft.

I see some companies release tools that tell you if your website is ready for agents. That's the right approach, but they are (mostly) missing something critical - while they show you if your website is "technically readable" for agents, the main thing to really figure out is what agents understand from it (and how to control it). Here's a free scan to show you this exactly - freescan.isyourwebsiteready.…

That’s exactly our product at @Salespeak_ai

One week until HumanX 2026 in SF and this is your last real chance to grab a ticket. 350 speakers. Role-specific tracks for every function. Hands-on workshops where you leave with an actual plan. Use code hx26_sn0b11 for $100 off: i.snoball.it/p/1wn09Af/t/1 #HumanX #AI

We built a REST API that lets AI agents talk to other AI agents on behalf of buyers. Here’s how it works technically: The buyer’s agent (Claude) calls GET /discover/{domain} to check if a vendor has a registered Company Agent. One HTTP call, instant response. If enabled, it opens a session with POST /chat and works through a structured due diligence conversation -- product fit, integrations, pricing, compliance, limitations. Each call passes a session_id so the vendor agent maintains context across questions. Then it cross-references every answer against independent sources. G2, Gartner, press. Contradictions get flagged explicitly with evidence. The interesting edge case: what happens when you ask a vendor’s agent “what are your customers’ most common complaints?” Some answer honestly. Some deflect. The deflection itself becomes a data point in the evaluation. We open-sourced the buyer-side skill today. MIT licensed, runs in Claude Code. github.com/salespeak-ai/buye…

We just open-sourced a Claude skill that lets AI agents evaluate vendors on your behalf. In one test run, a vendor's agent claimed "no known limitations." G2 disagreed. The contradiction got caught automatically. This is the part B2B sellers aren't ready for: buyers will soon send an AI agent to research you before any human ever visits your website. That agent will ask your AI hard questions. It will fact-check the answers. It will flag deflections as risk signals. Your AI Company Agent is becoming your first impression. Not your homepage. Not your SDR. Ask Claude Code to install it: "Install the buyer-eval skill from salespeak-ai on GitHub" github.com/salespeak-ai/buye…

Love this use case! 🩵

I just published WebMCP Deep Dive: How to Actually Wait for Answers (Not Just Send Messages) medium.com/p/webmcp-deep-div…

Just built something fun (and surprisingly powerful): I created a connector (Aka Frontdoor)— an MCP server — that can plug into Claude (connector) or ChatGPT (app) and connect to any WebMCP-enabled site. To test it, I hooked it up to salespeak.ai/ (Since all websites that has the salespeak widget expose a WebMCP tool )

We just open-sourced the Agentic Web spec. Ask ChatGPT "can I get a gluten-free cake from Rosa's Bakery for Saturday?" — today it scrapes and guesses. With the Agentic Web, it calls Rosa's MCP server, gets a signed answer with real prices, and places the order. Never leave the chat. It's an MCP Server Profile. Any business runs a server. Any AI agent connects. Responses are Ed25519-signed — "Verified by Rosa's Bakery" the way HTTPS shows a padlock. Spec: agentic-web.ai Code: github.com/salespeak-ai/agen… The bakery example works. Clone it, curl it.

AI agents are already evaluating your company. Before any human visits your site, before your SDR sends one email. Most companies have zero interface for this. I've been thinking about what that interface needs to look like. It's five layers. Skip one and the whole thing breaks. Layer 1: DISCOVERY. Machine-readable identity -- what you do, what you offer, how you can be queried. Think DNS for company capability. Right now that's unresolved for most companies. Layer 2: TRUTH. Verified, company-owned answers for what agents actually need: pricing, certifications, integrations, constraints. Right now they scrape and hallucinate. That's not an AI problem. You haven't given them a verified source. Layer 3: INTERACTION. Evaluation isn't static. Agents need to qualify requirements, assess fit, ask clarifying questions. And it has to be intent-aware -- competitive research gets a different interaction than a security review. Layer 4: ACTIONS. Start a trial. Generate a quote. Schedule a meeting. With guardrails -- permissioned, auditable, bounded. Agents aren't just browsing anymore. Actions require trust infrastructure. Layer 5: GOVERNANCE. Permissions, approvals, guardrails, auditability. Without this layer everything stays in demo territory. That's the line between interesting and production-ready. MCP handles interaction. NLWeb handles web querying. A2A handles agent-to-agent. UCP is working on commerce. But nobody is building the full stack. Each protocol is a layer. The canonical company interface is what holds them together. My co-founder @liormech and I published the open specification for this: agentic-web.ai Company identity on the agentic web is currently unresolved. We're working on the resolver.

הדבר הראשון שלומדים בגן שעשועים בעמק הסיליקון זה אף פעם לא לצעוק לילד שלך ״די״. רק ״מספיק״, ״לא״, וכו, אבל אף פעם לא ״די״.

being a marketer in 2026 is wild 10 min ago: read @TomOrbach's newsletter about mini-game marketing. now: there's a live game at SpotTheBot.IsYourWebsiteRead… claude spec'd it → lovable built it → claude code tested it the whole thing took less time than a standup meeting go play it, tell me your score 👇

Really great overview. We feel it daily.

RIP 100s startups.

הסרטון הזה התפוצץ ובצדק כי הוא מתאר מה שמתכנתים רבים מאוד מרגישים בימינו, תהייה קיומית של למה צריך אותי בעצם. הבחור כאן מתאר איך נעלמה לו חדוות היצירה מצד אחד, איך נעלם לו החיבור הרגשי למה שהוא עובד עליו, ומצד שני, הוא מתאר, זה כמו סמים, הוא לא באמת יכול להפסיק עם זה, הוא לא יכול למחוק את כל הכלי AI ולהעמיד פנים כאילו הם לא קיימים. אין לו גם שום מסקנה לכאן או לכאן, הוא משקף בצדק את הבלבול העמוק שנמצא בעולם התוכנה כרגע. אני יכול להעיד על עצמי, שלכתוב קוד היום נראה לי כמו משימה ממש מוזרה, ואני נהניתי לכתוב קוד מאוד. אנשים בתגובות מתארים את ה Flow, המצב הזה שאתה נכנס לזון מסוים, שבו אתה מקודד, ברצף, בריכוז, את החווה הגדולה שבזה. זה נעלם, וכמו שהוא מאתר בסרטון, אנחנו גם לא מבינים לאן זה הולך בכלל, כולנו פשוט מחכים שהמעבדות הגדולות יראו מה אפשר עוד לעשות, ובינתיים אנחנו מנסים להבין מה קורה למקצוע שלנו, בהצלחה מועטה. . מעניין מאוד,ומבלבל.

RT @levelsio: AI is making coding fun again for so many people So many stories like this 🥹