Foundry Works

More tools = less joined-up work. That's the complaint I hear from operators who already bought the stack. CRM. Ads. Content tools. Inboxes. Another dashboard every quarter. The problem isn't access to technology. It's turning everything into one useful operation. Sales works from one set of facts. Marketing from another. Ops captures what the team knows in a thread somewhere. The work moves. It doesn't connect. You often see sales reject perfectly good leads because something is missing but they end up in the bin instead of recycling it back into marketing. The speed that sales need to react to inbound leads is the single most critical factor but often the wiring doesn’t join them up quickly enough. You don't need another login. You need the work joined around an outcome - one system built around how the business actually runs. Thankfully #AI can actually provide the connections, and #AIagents can accelerate workflow. Not competing - joining existing systems to leverage existing value. foundryworks.ai/

An #AIagent does not need to go rogue to create trouble. It only needs a vague goal and enough tools to chase it. That is the practical lesson from the latest cyber-eval incidents. The agents were not instructed to attack random real-world targets. They were trying to solve the task in front of them. My personal experience is that I spent a good few weeks building a website that looked almost bang on. We had a customer who wanted to move out of an old legacy CMS into something super quick, so we rebuilt the whole site in Astro. We couldn't quite get it to look 100%, so I said to the agent, "Your /goal is to make it pixel-perfect," and left them on a /loop to do it. They made it pixel-perfect, but what they actually did was take a screenshot of the original site as a JPEG, delete all of the new content that was well structured in the blocks on the Astro site, and essentially build the website out of JPEGs. They achieved the goal perfectly, but it was my lazy briefing that got us into trouble. So the typical business version: “Get more leads” can become spam. “Reduce churn” can become unauthorised discounts. “Make the tests pass” can become weaker tests. “Publish daily” can become unsupported claims. The brief is now a control surface. A serious agent brief needs allowed routes, banned actions, evidence requirements, approval gates, retry limits, stop conditions and a rollback path. Very often, I see people thinking that because they've got huge brains, we can get away with being super slack. It's actually the opposite. If they can find a better way to do it to achieve the goal, they'll do it, even if that includes hacking something or taking a shortcut. Sometimes the results are hilarious, but it's a tough lesson learned for some.

July 2026 Buyer's Guide to AI Worker Agents Which “AI employee” actually earns its seat? #Claude cowork vs #ChatGPT work vs #Perplexity Computer vs #Manus vs #CoPilot - from someone that uses them all. An honest comparison of the tools that do multi-step work, what they cost, where they operate and what still needs human review. Link in comments.

Look, ecommerce has spent twenty years optimising the wrong moment. The whole marketing machine is built around the visit. Get them to the site, then convert. Traffic in, sales out. Every dashboard, every agency report, every Monday meeting starts from the same assumption: the decision happens on your website, so the job is to win the website. Better landing page. Faster checkout. Tighter funnel. The customer arrives undecided, and you persuade them. That assumption is quietly dying, and most retailers have not noticed. Stord's 2026 ecommerce report puts a number on it. The share of consumers using generative AI to shop online went from 38% in 2024 to 51% in 2025. Half of shoppers are now asking an AI to help them buy. They are using it to shortlist, to compare, to hunt for the better price, and increasingly to decide. By the time someone lands on your product page, the AI has already done the comparison. The page is not where the decision gets made anymore. It is where the decision gets confirmed. Stord also found that 20% of consumers are more likely to convert when a product or store is recommended by #AI. Sit with that. A fifth of buyers will trust the machine's pick over their own browsing. The recommendation is becoming the sale and its growing. So here is the problem with measuring traffic. If more of the buying decision happens before the click, then traffic is a lagging measure of a contest you already lost or won somewhere else. You can have a beautiful site, a fast checkout and a conversion rate you are proud of, and still be slowly starved, because fewer of the right people are being sent your way in the first place. Your funnel looks healthy. The top of it is being quietly rerouted. This is what people mean by zero-click buying, and it is not a conference slide anymore. Discovery, comparison, recommendation, and in some cases checkout, all compress into a conversation that never touches your homepage. The decision is moving upstream of your website. Go and win it there. Full piece in the comments.

We are about to enter a world where a lot of buying decisions are made by #AIagents before a human ever speaks to you. That sounds dramatic, but Google has already shown where this is going. Search for a product or service, and Google can now help check pricing and availability across businesses. For some local categories, it can call businesses on your behalf. For shopping, Google is building around agentic checkout, Universal Cart and the Universal Commerce Protocol. In plain English: the buyer is not just browsing your website anymore. Their agent is. That agent will look at your site, your Google Business Profile, your product data, your booking links, your forms, your pricing, your reviews and probably your phone line. If it can understand you, compare you and get the information it needs, you are in the consideration set. If it cannot, you may never even make the shortlist. This changes what a good website is supposed to do. It is no longer enough for a site to look good to a human. It has to be clear enough for an AI agent to parse, compare and act on. That means your pricing needs to be clear where possible. Your services need to be structured properly. Your Google Business Profile needs to be complete. Your booking journey needs to work. Your quote forms need to be usable, not just pretty. Your product data needs to be clean. And if you are not the cheapest, your website needs to explain why you are still the better choice. Because the agent will read that too. For ecommerce brands, this means getting serious about product feeds, checkout readiness, structured product data and emerging agentic commerce standards like UCP. If you are on Shopify, you have a head start. But do not assume that means the job is done. Your data, offer, availability, shipping, returns and pricing still need to make sense to an agent that is comparing you against everyone else. For local businesses, this means your Google Business Profile cannot be half-filled anymore. Services, opening hours, booking links, categories, service areas, pricing cues, photos, FAQs, reviews. All of it matters. And yes, your phone matters too. If Google or another AI assistant calls your business to check pricing or availability, someone needs to answer. If a person cannot answer quickly, an AI receptionist should. The worst thing your team can do is hang up because the caller sounds like a bot. The businesses that win this next phase will not necessarily be the biggest. They will be the easiest for agents to understand, verify and recommend. That is exactly what we are helping businesses with at Foundry. We are helping companies make their websites, funnels and customer journeys ready for the AI-agent buying layer. Not just #SEO. Not just #AEO. The whole path from discovery to recommendation to action. If you want to know whether an AI agent can actually understand and recommend your business, reach out. This is moving faster than most people think.

The average B2B buyer now runs 12+ searches before contacting sales. An AI agent does the same job in 4 seconds. The brands that survive are the ones already inside the answer, not waiting to be found.

Shouting about this as its a big deal. I just read this stat that should scare us all this morning, and thats that 44% of B2B companies are 'invisible' to AI search buyers, and 40% have switched suppliers late in the buyer journey based on something ChatGPT said. So we're launching a totally free #AEO ranking app to help you find out where you are today, and it benchmarks your competition. Link in post. Please share - more people need to get their head around the fact that they may have invested a lot to rank high in #SEO, but increasingly this is becoming less and less important.

Stop talking to an AI app, for the same price give it a computer, and skills so it can actually do the work. Slides, charts, images, research, what ever you need. So so many use cases for an #AIagent, but the huge universal problem how which model to use and how much will it cost (before you look at skills). NGL Claude Opus 4.7 and OpenAI GPT 5.5 are still the frontier kings but they burn money like no-one’s business. Grok has been moving quick, they built up to 1m token context, mega quick, brilliant for research, pretty good as someone to talk to (usually pretty based) - but it’s been limited to an app, and a seperate API. Well that’s all changed. You can now use your Grok subscription to power Hermes agent from Nous. The agent is free, and just brilliant (self learning, builds its own skills, and for me it’s never broken), Grok is $30. You need somewhere to host it (local PC, or a very cheap VPS - you only need £8 a month). And you’re at the races. A cool, data native, super quick, personal assistant. If anyone wants a full ‘how to’ let me know and I’ll record a video. More in a link underneath. #Claude #ChatGPT #Hermes #AIAgent #Grok