<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Paul Lovelace on Medium]]></title>
        <description><![CDATA[Stories by Paul Lovelace on Medium]]></description>
        <link>https://medium.com/@paullovelace?source=rss-112fd1c4035e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Stories by Paul Lovelace on Medium</title>
            <link>https://medium.com/@paullovelace?source=rss-112fd1c4035e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 05:08:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@paullovelace/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Google Just Gave Its AI a Work Email. Yes, Really.]]></title>
            <link>https://medium.com/@paullovelace/google-just-gave-its-ai-a-work-email-yes-really-3d77d536db05?source=rss-112fd1c4035e------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d77d536db05</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[google]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Paul Lovelace]]></dc:creator>
            <pubDate>Tue, 13 Oct 2026 09:29:34 GMT</pubDate>
            <atom:updated>2026-10-13T09:29:34.049Z</atom:updated>
            <content:encoded><![CDATA[<p>Google announced something last week that sounds like a joke until you think about it for thirty seconds: its new AI assistant gets its own company email address.</p><p>It’s called the Gemini agent. It was unveiled on October 8 at Google’s “Gemini at Work” event. And the email address is the least weird part.</p><h3>What it actually is</h3><p>Forget chatbots. The Gemini agent is Google’s attempt at a “universal agent for work” — you give it an objective, not step-by-step instructions, and it goes off and does the job.</p><p>“Prepare the quarterly report” — and it plans the work, pulls data, writes the draft, and brings back something finished. It keeps working after you close your laptop. Some jobs run for hours. Some for days.</p><p>This isn’t new as an idea. What’s new is how seriously Google is treating the agent as a *colleague* rather than a tool.</p><h3>The part that made everyone pause: it gets hired</h3><p>Here’s the genuinely strange bit. A company can create a persistent “coworker agent” — say, an events coordinator — and Google gives it:</p><ul><li>Its own **email address** (something@agents.company.com)</li><li>Its own **calendar**</li><li>Its own **file storage**</li><li>A spot in the **company directory**</li></ul><p>Colleagues add it to group chats. They @mention it in documents. Its edits show up in version history under its own name — not yours.</p><p>Bloomberg’s write-up nailed why this matters: the industry has stopped treating agents as a feature and started treating them as something close to headcount. When the AI has an email address and a calendar, the line between “tool” and “teammate” gets very blurry.</p><h3>What it can actually touch</h3><p>The connector list reads like an IT department’s inventory: Gmail, Docs, Drive, Sheets, Calendar, Slack, Microsoft 365, Jira, Confluence, Git, BigQuery, Databricks, Postgres, Snowflake, Salesforce, ServiceNow — plus anything speaking the MCP protocol.</p><p>It doesn’t just use Google’s models, either. It routes each job to whatever fits best — Google’s Gemini models today, Anthropic’s Claude too, with OpenAI and open models promised later. A Google product that will happily hand your work to a competitor’s model is… notable.</p><h3>The safety brakes (and why they exist)</h3><p>Google clearly knows this is alarming, because the announcement led with governance:</p><ul><li>Every action is logged under the **agent’s own identity** — cryptographically attested, not borrowed from a human</li><li>Companies set **hard spending limits** — hit the cap and the agent stops until a human approves more</li><li>It only sees **what you share with it**, per Google</li><li>Role-based permissions, audit trails, an isolated sandbox for agent activity</li></ul><p>The emphasis isn’t accidental. Lawmakers recently cited a $12.9 billion hack in a proposed AI liability bill. An agent with days-long autonomy and access to your Gmail, Drive, and Slack is a bigger attack surface than a chatbot that answers questions. Google gated the whole thing behind a private preview — no public launch — which tells you how seriously they take the blast radius question.</p><p>Worth noting what wasn’t in the announcement: no specific security certifications, no compliance frameworks detailed. Enterprises will want those answers before GA.</p><h3>The money question nobody can answer</h3><p>Google hasn’t announced pricing. A company rep said there are no plans for per-seat licensing of coworker agents in Workspace — which is interesting, because per-seat is how enterprise software has been sold for decades. If agents aren’t seats, what are they? Metered compute? A new pricing model entirely? Undisclosed for now.</p><p>Early testers include On, Shopify, and PayPal. General availability is expected around end of October, starting in North America. For context on scale: Google says Gemini has over a billion monthly users and nearly 90% of Fortune 100 companies use Gemini Enterprise.</p><h3>Who should care</h3><ul><li>**Anyone managing a team** — the “coworker with an email address” model forces real questions: who reviews the agent’s work? Who’s accountable when it’s wrong? What does onboarding an AI even mean?</li><li>**IT and security people** — identity management now includes non-humans with mailboxes. Your access-review process just got more complicated.</li><li>**Everyone else watching AI** — this is the clearest signal yet of where the industry is going: not smarter chatbots, but AI that holds a job title.</li></ul><h3>Who shouldn’t care (yet)</h3><p>If your company isn’t on Google Workspace or deep in Microsoft 365, this is a spectator sport for now. Private preview means most of us won’t touch it for months. And the pricing unknown makes ROI math impossible.</p><h3>Bottom line</h3><p>Google gave its AI an email address because that’s what it takes to make an agent a real coworker: identity, persistence, accountability. The technology is impressive. The governance story is serious. The unanswered questions — price, certifications, what happens when it goes wrong — are the ones that will decide whether this is the future of work or an expensive experiment.</p><p>The hype says your next colleague is AI. The reality says: maybe, but read the audit log first.</p><p>*Sources: Google Cloud announcement (Oct 8, 2026), TechCrunch, Bloomberg, 9to5Google, Futurum Group. Availability and pricing details were undisclosed at announcement; treat timelines as Google’s stated plans.*</p><p><em>Originally published at </em><a href="https://muse.ai/files/1342912955576372/2411015276392737/7fipwryff1g7xg1zh07b9oaj/article2-gemini-agent.html"><em>https://muse.ai</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d77d536db05" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Muse Wonder Explained in Plain English: Meta’s New AI, What It Costs, and Who Should Care]]></title>
            <link>https://medium.com/@paullovelace/muse-wonder-explained-in-plain-english-metas-new-ai-what-it-costs-and-who-should-care-67dfa9e03e01?source=rss-112fd1c4035e------2</link>
            <guid isPermaLink="false">https://medium.com/p/67dfa9e03e01</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[meta]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Paul Lovelace]]></dc:creator>
            <pubDate>Tue, 13 Oct 2026 09:26:44 GMT</pubDate>
            <atom:updated>2026-10-13T09:26:44.079Z</atom:updated>
            <content:encoded><![CDATA[<h3>What actually launched</h3><p>Two models, and the difference matters:</p><ul><li>**Muse Wonder** — available through Meta’s API right now. This is the one developers can actually use.</li><li>**Muse Wonder Full** — the stronger version, but you **can’t** get it through the API. It’s locked behind paid Muse Code subscriptions.</li></ul><p>So the best version is subscription-only. Remember that — it comes up again.</p><h3>The real story: Meta’s cheap era is over</h3><p>For years, Meta was the “affordable” AI company. Llama models were cheap or free. Developers loved it.</p><p>Wonder kills that era. The price: <strong>$4 per million input tokens, $20 per million output tokens</strong> — exactly what Anthropic charges for Claude Opus 5.5. Same numbers, down to the cent.</p><p>Meta spent two years saying it would stay open and affordable while others charged premium prices. Wonder’s price tag says that era is finished.</p><h3>What it’s actually good at</h3><p>Forget the marketing. Independent testers (Artificial Analysis) put Wonder Full at <strong>56</strong> on their Intelligence Index — behind Claude Opus 5.5 (58), ahead of GPT-6 Astra (53). Solid, not dominant.</p><p>Where it genuinely shines:</p><ul><li>**Memory.** Meta’s big bet is “Active Memory” — the model remembers context across long conversations. On memory tests it beat every competitor.</li><li>**Marathon coding.** On ultra-long software projects (SWE-Marathon), it scored 57.5 vs 46.3 for Opus. If you need AI to work on something for hours, this matters.</li><li>**Agent work.** Legal document tasks, business workflows — it leads the pack.</li></ul><p>Where it’s weak:</p><h3>The controversy</h3><p>Not everyone buys Meta’s numbers. Bindu Reddy, CEO of Abacus.AI (a company that tests AI models for a living), publicly disputed the magnitude of Meta’s performance claims — expect more independent scrutiny in the coming weeks.</p><p>Meta’s own benchmarks show Wonder losing 9 of its own 14 tests to competitors. Read that again: *Meta’s own chosen tests*, and it still loses most of them.</p><h3>Who should care</h3><ul><li>**Developers** choosing between Wonder and Opus 5.5 at identical prices — the memory and long-task strengths might tip it.</li><li>**Anyone watching Meta** — this is the company’s first serious attempt at a frontier model, and the pricing signals a totally new Meta.</li><li>**AI skeptics** — the gap between “frontier” marketing and benchmark reality is worth watching.</li></ul><h3>Who shouldn’t care</h3><p>Everyone else, honestly. If you’re not building with AI or choosing between models, this changes nothing about your daily life today. The models will show up inside Meta’s products eventually — that’s when it matters for regular people.</p><h3>Bottom line</h3><p>Muse Wonder is a real frontier model from a company that finally decided to play the premium game. It’s genuinely good at memory and long tasks, genuinely weak at general knowledge, and priced exactly like its biggest rival. The hype says revolution. The numbers say: solid second place with one great trick.</p><p>— -</p><p>*Sources: TechCrunch, VentureBeat, Artificial Analysis Intelligence Index, Meta model card. Benchmark numbers are from independent testing where noted.*</p><p><em>Originally published at </em><a href="https://muse.ai/files/1342912955576372/1866888744658426/72su6dw1nrpw02dftc244kgl/article1-muse-wonder.html"><em>https://muse.ai</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=67dfa9e03e01" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>