[{"data":1,"prerenderedAt":1146},["ShallowReactive",2],{"market-markets_en-dubai":3},{"market":4,"cases":125,"notes":472},{"id":5,"title":6,"areaServed":7,"body":11,"description":15,"draft":18,"explore":19,"extension":32,"eyebrow":33,"facts":34,"faq":47,"faqTitle":66,"heroNote":67,"how":68,"howTitle":85,"hue":86,"meta":87,"name":88,"navigation":89,"offerings":90,"offeringsTitle":106,"path":107,"proof":108,"related":115,"seo":119,"seoDescription":120,"seoTitle":121,"stem":122,"summary":123,"__hash__":124},"markets_en\u002Fdubai.md","The AI agency in Dubai that ==ships in 30 days.==",[8,9,10],"Dubai","United Arab Emirates","Middle East",{"type":12,"value":13,"toc":14},"minimark",[],{"title":15,"searchDepth":16,"depth":16,"links":17},"",2,[],false,[20,24,28],{"label":21,"title":22,"href":23},"Use cases","Twelve jobs agents do every day, and the cases where they run.","\u002Fcases#use-cases",{"label":25,"title":26,"href":27},"Agentic OS","What an Agentic OS is, and how it runs inside your tools.","\u002Fagentic-os",{"label":29,"title":30,"href":31},"Engineering","Custom software, built around how you work.","\u002Fengineering\u002Fcustom-software","md","Dubai · United Arab Emirates",[35,38,41,44],{"label":36,"value":37},"Office","Dubai, GMT+4, with a second office in Cascais",{"label":39,"value":40},"Built for","Mid-sized and large companies in the UAE",{"label":42,"value":43},"Live in","30 days, or your money back",{"label":45,"value":46},"Hosted","In your cloud, in a UAE region if needed",[48,51,54,57,60,63],{"q":49,"a":50},"Do you have an office in Dubai?","Yes. WizardingCode works from Dubai and from Cascais, Portugal. Diagnostic workshops and launches can happen on site in Dubai, and the build runs remotely, in English, so your team only gives the hours that matter.",{"q":52,"a":53},"Which companies in Dubai is this for?","Companies with volume and repetition: real estate agencies and developers, hospitality, retail and e-commerce, logistics and professional services. If a team spends its day answering the same questions or copying data between tools, there is an agent for that work.",{"q":55,"a":56},"Can the agents answer customers in Arabic?","Yes. The agents answer in the language the customer writes in, Arabic and English included, and a second model checks every answer against your policies before it goes out. Your team decides which conversations always go to a person.",{"q":58,"a":59},"Where is our data stored?","Where you choose. The agents can run in your own cloud account, in a UAE region if your data has to stay in the country, or in ours. Each agent only gets the access it needs, every step is logged, and your data is never used to train models. We document which personal data each agent touches, so your legal team can check it against the UAE data protection law, or the DIFC or ADGM rules if you are in those free zones.",{"q":61,"a":62},"How long does it take, and what does it cost?","Live in 30 days, or we refund the setup fee. The price is fixed after a one-week Diagnostic, because it depends on how many agents, which systems and how much data we have to organise. The first call is free and takes 30 minutes.",{"q":64,"a":65},"Do you also build custom software in Dubai?","Yes. Before agents, we spent years building web platforms, mobile apps, back-offices and integrations, and we still do, quoted after a short discovery. Everything we build is ready for agents from day one.","What Dubai companies ask us first.","Fixed price after a one-week Diagnostic",[69,73,77,81],{"step":70,"title":71,"detail":72},"Week 0","Diagnose","A free call, then one week of workshops with your teams, in Dubai or remotely. Every job ranked by value, and a fixed price.",{"step":74,"title":75,"detail":76},"Week 1–2","Map and build","Access to your CRM, WhatsApp and the tools you use, real cases and the success metric signed off. Then the agents.",{"step":78,"title":79,"detail":80},"Week 3","Prove","Tests on your real conversations and documents, then shadow mode with your team before anything reaches a client.",{"step":82,"title":83,"detail":84},"Day 30","Live","In production, monitored, team trained. Then we run the agents and add the next one as the numbers come in.","From the first meeting to live agents ==in 30 days.==","sun",{},"AI agency in Dubai",true,[91,94,98,102],{"title":92,"detail":93,"hue":86},"Leads answered on WhatsApp","Enquiries from portals, ads and WhatsApp answered in minutes, day and night, in the language the customer writes in. Qualified leads go straight to the right person.",{"title":95,"detail":96,"hue":97},"Market and listing watch","Portals, registries and competitor sites read every hour. Every new listing, tender or price change scored and sent to your team.","magenta",{"title":99,"detail":100,"hue":101},"Inboxes and back-office","Shared inboxes sorted and answered, documents read, and bookings and orders entered into your CRM or ERP without copy and paste.","violet",{"title":103,"detail":104,"hue":105},"Reports on demand","The weekly pipeline, sales or operations report written from your own tools, before the meeting where someone asks for it.","blue","What agents do for Dubai companies","\u002Fdubai",{"title":109,"lede":110,"cases":111,"note":114},"In Dubai, half a day ~~is enough to lose a lead.~~","Speed decides who signs the owner and who books the viewing. Our Dubai real estate case shows what changes when a Scraper reads the market every hour and a Responder answers every enquiry in minutes.",[112,113],"industry-tenders","ecommerce-marketplace","The real estate case is illustrative, based on real projects. Its figures are representative.",[116,117,118],"the-30-day-playbook-week-by-week","what-agents-should-never-do-alone","why-95-percent-of-ai-pilots-never-reach-production",{"title":6,"description":15},"An AI agency with an office in Dubai. We build AI agents that answer leads on WhatsApp, sort inboxes and write reports for UAE companies, live in 30 days.","AI agency in Dubai for agents and automation","dubai","WizardingCode builds and runs AI agents for companies in Dubai and across the UAE. Agents that answer leads on WhatsApp in minutes, sort shared inboxes, watch the market and write the weekly report, inside the tools you already use and under rules your people set. Live in 30 days, or your money back.","B2F6jf-Qzohmk6zEnLAigka8ZI1FbOJhSsYAMmwcrrI",[126,256],{"id":127,"title":128,"after":129,"agents":157,"before":161,"body":166,"built":170,"capabilities":198,"client":199,"description":15,"draft":18,"extension":32,"facts":200,"faq":198,"faqTitle":198,"flagship":18,"hue":105,"industry":212,"industryLabel":203,"meta":213,"metrics":214,"metricsNote":224,"moreCases":225,"moreRow":228,"navigation":89,"path":231,"related":198,"resultQuote":198,"seo":232,"seoDescription":198,"seoTitle":198,"stem":233,"summary":234,"teaser":235,"timeline":237,"__hash__":255},"cases_en\u002Fcases\u002Findustry-tenders.md","Every listing found ==before the competition.==",{"title":130,"lede":131,"rows":132},"What changed.","Measured from the command center, against the metric agreed in week one.",[133,137,141,145,149,153],{"metric":134,"before":135,"after":136},"Time to spot a new listing","2 days","20 min",{"metric":138,"before":139,"after":140},"Leads answered within 1 hour","30%","95%",{"metric":142,"before":143,"after":144},"Listings tracked","200","3,000+",{"metric":146,"before":147,"after":148},"Weekly pipeline report","Half a day","On demand",{"metric":150,"before":151,"after":152},"Viewings booked per month","40","110",{"metric":154,"before":155,"after":156},"Exclusive listings signed per quarter","5","14",[158,159,160],"Scraper","Analyst","Responder",{"title":162,"paragraphs":163},"The best listings went to whoever called first.",[164,165],"New listings appear every hour on the property portals, in developer releases and in public registries. Two junior brokers checked them by hand each morning and pasted the good ones into a spreadsheet.","By the time a broker called, a faster agency had often signed the owner. Buyer enquiries waited half a day for a reply. In Dubai, half a day is enough to lose them.",{"type":12,"value":167,"toc":168},[],{"title":15,"searchDepth":16,"depth":16,"links":169},[],{"title":171,"lede":172,"nodes":173,"approvalRule":195,"sharedMemory":196,"connectedTo":197},"A sales floor that ~~reads the market every hour.~~","Brokers get new listings and scored leads in WhatsApp and the CRM. The weekly pipeline report writes itself.",[174,179,183,187,190],{"glyph":175,"label":176,"detail":177,"hue":105,"kind":178},"@","Portals and registries","Property portals, developer releases and public registries, checked every 15 minutes.","system",{"glyph":180,"label":158,"detail":181,"hue":105,"kind":182},"S","Finds every new listing and price change, and pulls out area, size, price and agent.","agent",{"glyph":184,"label":159,"detail":185,"hue":186,"kind":182},"A","Scores leads and listings against recent sales, suggests a price and writes the weekly report.","aqua",{"glyph":188,"label":160,"detail":189,"hue":97,"kind":182},"R","Answers new buyer and owner enquiries on WhatsApp in minutes and books the viewing.",{"glyph":191,"label":192,"detail":193,"hue":86,"kind":194},"!","Brokers","Pricing advice, offers and anything signed go to a broker, with the full history.","team","No price, offer or contract goes out without a broker","Past deals, price history by area, listings, client history","Property portals · CRM · WhatsApp · Google Sheets",null,"Real estate agency, Dubai",[201,204,207,210],{"label":202,"value":203},"Industry","Real estate",{"label":205,"value":206},"Team","Sales and listings",{"label":208,"value":209},"Agents","Scraper + Analyst + Responder",{"label":42,"value":211},"30 days","realestate",{},[215,218,221],{"value":216,"label":217},"€340k","in commission from new leads, year one",{"value":219,"label":220},"€35k","one-off build",{"value":222,"label":223},"\u003C 3 mo","payback","Illustrative case based on real projects. Figures are representative.",[113,226,227],"logistics-inbox","retail-customer-care",{"title":229,"agents":230},"Every listing found before the competition.","Scraper · Analyst · Responder","\u002Fcases\u002Findustry-tenders",{"title":128,"description":15},"cases\u002Findustry-tenders","A Dubai agency kept losing listings and buyers to faster competitors. Now a Scraper watches every portal and registry, an Analyst scores what it finds, and a Responder answers new leads in minutes.",{"value":216,"label":236,"title":229},"in commission from new leads",[238,241,245,249,252],{"week":239,"name":71,"detail":240},"WEEK 0","Two days with the brokers. Mapped where listings appear and where leads were lost.",{"week":242,"name":243,"detail":244},"WEEK 1","Map","Access to the CRM, the portals and WhatsApp Business. Metric signed off: time to spot a new listing.",{"week":246,"name":247,"detail":248},"WEEK 2","Build","Scraper, Analyst and Responder built, scoring tuned with the head of sales.",{"week":250,"name":79,"detail":251},"WEEK 3","Two weeks side by side with the manual check. The agents tracked 15× more listings.",{"week":253,"name":83,"detail":254},"DAY 30","Alerts live in WhatsApp as listings appear. Brokers trained on the command center.","byE-lEG_l2E_yKhTVmAhn0QHaCJ61Ndpy6zTC0YLygQ",{"id":257,"title":258,"after":259,"agents":278,"before":281,"body":286,"built":290,"capabilities":310,"client":198,"description":15,"draft":18,"extension":32,"facts":402,"faq":412,"faqTitle":430,"flagship":89,"hue":101,"industry":431,"industryLabel":432,"meta":433,"metrics":434,"metricsNote":198,"moreCases":444,"moreRow":446,"navigation":89,"path":448,"related":449,"resultQuote":198,"seo":453,"seoDescription":454,"seoTitle":455,"stem":456,"summary":457,"teaser":458,"timeline":460,"__hash__":471},"cases_en\u002Fcases\u002Fecommerce-marketplace.md","One marketplace. 90 agents. ==Every team judged.==",{"title":130,"lede":131,"rows":260},[261,265,268,272,276],{"metric":262,"before":263,"after":264},"Supplier import cleaned and matched","~2 days","~20 min",{"metric":266,"before":264,"after":267},"Product page, written and translated","~1 min, reviewed",{"metric":269,"before":270,"after":271},"Support tickets with a ready draft","0%","~70%",{"metric":273,"before":274,"after":275},"Stalled stock found and priced for clearance","Monthly, by hand","Daily report",{"metric":277,"before":147,"after":148},"Weekly sales report",[279,280],"Full Agentic OS","8 departments",{"title":282,"paragraphs":283},"Thousands of products. A handful of people.",[284,285],"Every flash sale meant supplier files to clean, product pages to write in several languages, prices to check against margins and stock to reconcile with the warehouse. Then the tickets, the refunds and the campaign reports.","The team was great. There just weren’t enough hours, and the work that made money, like pricing and recovering stalled stock, always came last.",{"type":12,"value":287,"toc":288},[],{"title":15,"searchDepth":16,"depth":16,"links":289},[],{"title":291,"lede":292,"nodes":293,"approvalRule":307,"sharedMemory":308,"connectedTo":309},"A full Agentic OS, ~~inside their own CRM.~~","Eight departments, from commerce to wallet, each with lead agents, specialists and a judge.",[294,297,301,304],{"glyph":175,"label":295,"detail":296,"hue":105,"kind":178},"The CRM","Orders, catalogue, suppliers, tickets, campaigns and wallet in one back-office.",{"glyph":298,"label":299,"detail":300,"hue":86,"kind":182},"O","Jarvis, the orchestrator","One chat for the whole company. Understands the ask and runs the right teams in parallel.",{"glyph":302,"label":280,"detail":303,"hue":101,"kind":182},"8","Commerce, support, marketing, operations, analytics, wallet, catalogue, platform.",{"glyph":191,"label":305,"detail":306,"hue":86,"kind":194},"Judges + your team","A judge on a different model checks every team’s answer. Money and order changes always go to a person.","Refunds, prices and order changes are never executed without a person","Catalogue, margins, policies, customer history, twelve months of tickets","Their CRM · storefront · ERP · Google and Meta Ads · email, SMS, push, WhatsApp · payments · couriers",{"title":311,"lede":312,"groups":313},"Everything a growing store ==wishes it had.==","Twenty-six capabilities, all in one OS. Every one of them can be built for your business.",[314,345,375],{"group":315,"items":316},"Sell more",[317,320,323,326,330,333,336,339,342],{"glyph":188,"name":318,"detail":319,"hue":97},"Cart recovery","Brings abandoned carts back, and proposes each week how to recover more, measured 28 days later.",{"glyph":184,"name":321,"detail":322,"hue":186},"Customer insights","Persona, churn risk and next best action for every buyer, refreshed every night.",{"glyph":184,"name":324,"detail":325,"hue":186},"Synthetic customers","Two AI persona models predict opens, clicks and unsubscribes before a send, and can block a launch.",{"glyph":327,"name":328,"detail":329,"hue":101},"C","Segments in plain words","Describe an audience, get the rules and the real count. New segments are found and named daily.",{"glyph":188,"name":331,"detail":332,"hue":97},"Campaign planner","One brief becomes an email, SMS, push and Google Ads plan, with audience and cost.",{"glyph":188,"name":334,"detail":335,"hue":97},"Segment sends","Newsletter, SMS and push studios write for each segment, in four languages.",{"glyph":180,"name":337,"detail":338,"hue":105},"Ads assistants","Read Google and Meta Ads, find wasted spend, propose what to pause, boost or scale.",{"glyph":188,"name":340,"detail":341,"hue":97},"Social studio","Weekly Facebook and Instagram plan, captions and hashtags, inbox triage with reply drafts.",{"glyph":184,"name":343,"detail":344,"hue":186},"Sales reaction","Reads sales as they move and adapts campaigns, offers and stock pushes the same day.",{"group":346,"items":347},"Protect margin",[348,351,354,357,360,363,366,369,372],{"glyph":327,"name":349,"detail":350,"hue":101},"Margin guard","A hard veto on any price or offer below the floor, after returns, fees and shipping. No agent writes a price.",{"glyph":184,"name":352,"detail":353,"hue":186},"Return oracle","Measured return rates and the expected return cost per product, inside every margin calculation.",{"glyph":184,"name":355,"detail":356,"hue":186},"Dead stock recovery","Finds stock with no sale in 60 days and drafts the discount, with copy that quotes the breakeven.",{"glyph":184,"name":358,"detail":359,"hue":186},"Warehouse watch","Daily stalled-stock value, articles invisible to the store and divergences, grouped into clearance proposals.",{"glyph":327,"name":361,"detail":362,"hue":101},"Wallet fraud review","Fourteen audit rules and an AI team trace every suspicious credit, refund and balance, nightly.",{"glyph":184,"name":364,"detail":365,"hue":186},"Loss reasons","Mines tickets, returns and failed recoveries into cited reasons why sales are lost.",{"glyph":327,"name":367,"detail":368,"hue":101},"Supplier price check","Judges every dropship price against your pricing rules, every hour.",{"glyph":184,"name":370,"detail":371,"hue":186},"Daily action brief","Every morning at six: what moved yesterday, why, and what to do about it.",{"glyph":180,"name":373,"detail":374,"hue":105},"Competitor analysis","Tracks competitor prices, launches and promotions and says where you are exposed.",{"group":376,"items":377},"Run the back-office",[378,381,384,387,390,393,396,399],{"glyph":298,"name":379,"detail":380,"hue":86},"Jarvis, one chat","Any business question, routed to six teams in parallel, every answer judged.",{"glyph":188,"name":382,"detail":383,"hue":97},"Support replies","Triage, refund and policy answers, and drafts that send themselves when the judge allows.",{"glyph":327,"name":385,"detail":386,"hue":101},"Supplier imports","Messy supplier files resolved to categories, attributes and products, re-checked by a judge.",{"glyph":188,"name":388,"detail":389,"hue":97},"Product studio","Product, category, brand and page copy with SEO and AI-search content, in four languages.",{"glyph":180,"name":391,"detail":392,"hue":105},"Logistics","Courier diagnosis, shipping rules per market, tracking and labels.",{"glyph":327,"name":394,"detail":395,"hue":101},"Company memory","FAQs, policies, twelve months of ticket resolutions and documents, searchable, with contradiction alerts.",{"glyph":298,"name":397,"detail":398,"hue":86},"Lessons with a human gate","Judge and human feedback distilled nightly into lessons a person approves before agents use them.",{"glyph":184,"name":400,"detail":401,"hue":186},"Smart reports","Ask in plain words, get charts, tables and commentary, on a schedule.",[403,405,408,410],{"label":202,"value":404},"Fashion e-commerce",{"label":406,"value":407},"Scope","Full back-office",{"label":208,"value":409},"90, in 8 departments",{"label":42,"value":411},"First squad in 30 days",[413,416,419,422,425,428],{"q":414,"a":415},"What can AI agents do for a fashion e-commerce business?","Here, they take the work that repeats with every flash sale: cleaning supplier files, writing and translating product pages, answering support tickets, recovering abandoned carts, checking prices against margins and finding stock to clear before it goes stale. The team keeps the buying, the brand and every decision that moves money.",{"q":417,"a":418},"Can the agents write product pages in several languages?","Yes. In this marketplace a product page, written and translated, went from about 20 minutes to about one, with a person reviewing it. The product studio writes in four languages, from the product data and the brand’s tone of voice.",{"q":420,"a":421},"Does this work on Shopify or Shopify Plus?","Yes. This marketplace runs its agents inside its own CRM, but the same agents connect to Shopify and Shopify Plus through the Admin API and webhooks, or to a custom platform. They work where your team works today, so nothing has to be migrated.",{"q":423,"a":424},"Can an agent give discounts or refunds on its own?","No. Refunds, prices and order changes are never executed without a person. A margin guard, written as plain code rather than a model, vetoes any price or offer below the floor of each product, after returns, fees and shipping.",{"q":426,"a":427},"What if an agent gets a size or a policy wrong?","Every team has a judge: a second, different model that checks each answer against the catalogue, the policies and the company memory before it goes out. When the facts do not match, the answer goes to a person instead.",{"q":61,"a":429},"The first squad goes live in 30 days, or we refund the setup fee. The price is fixed after a one-week Diagnostic, because it depends on how many agents, which systems and how much data we have to organise. The first call is free and takes 30 minutes.","What fashion brands ask us first.","ecommerce","E-commerce marketplace",{},[435,438,441],{"value":436,"label":437},"90","agents in production",{"value":439,"label":440},"440+","tools they can use",{"value":442,"label":443},"9","judges checking their work",[226,227,445],"services-reporting",{"title":447,"agents":409},"90 agents run a fashion marketplace’s back-office.","\u002Fcases\u002Fecommerce-marketplace",[450,451,452],"cart-recovery-with-agents-notes-from-a-fashion-marketplace","testing-a-campaign-on-customers-who-dont-exist","every-agent-needs-a-judge",{"title":258,"description":15},"A fashion flash-sales marketplace runs 90 AI agents in eight departments: support, cart recovery, product pages, pricing and stock. First squad live in 30 days.","AI agents for fashion e-commerce: 90 in production","cases\u002Fecommerce-marketplace","A fashion flash-sales marketplace moves thousands of products a week across suppliers, campaigns, orders and support. We built its Agentic OS inside the CRM: 90 agents in eight departments, one orchestrator the whole team talks to, and a judge on every team that checks the answers.",{"value":436,"label":437,"title":459},"One marketplace. 90 agents. Every team judged.",[461,463,465,467,469],{"week":239,"name":71,"detail":462},"Mapped the back-office with each team: where the hours go, where the money leaks.",{"week":242,"name":243,"detail":464},"Access to the CRM and data. First metric signed off: support first response time.",{"week":246,"name":247,"detail":466},"Support squad built: triage, reply drafter, refunds specialist and a judge.",{"week":250,"name":79,"detail":468},"Shadow mode on live tickets, with handoff rules the team could tune.",{"week":253,"name":83,"detail":470},"Support squad live. Then one new squad a month: catalogue, pricing, marketing, reports.","TszbeOA7d2tekEJh_FQ6SdcQm2mc8rFETn-2iPt8c-A",[473,695,936],{"id":474,"title":475,"author":476,"blobHue":198,"body":480,"category":675,"date":676,"description":485,"draft":18,"extension":32,"featured":18,"headline":677,"hue":97,"letter":678,"meta":679,"navigation":89,"path":680,"readMinutes":681,"related":682,"seo":684,"stem":685,"summary":686,"toc":687,"__hash__":694},"insights_en\u002Finsights\u002Fthe-30-day-playbook-week-by-week.md","The 30-day playbook, week by week",{"name":477,"role":478,"bio":479},"André","Founder & CTO","André is the founder and CTO of WizardingCode. Eight years building the software companies run on, now putting agents into production.",{"type":12,"value":481,"toc":667},[482,486,489,494,497,500,504,507,536,539,543,546,549,552,556,559,562,647,651,654,657,661,664],[483,484,485],"p",{},"Thirty days sounds fast for putting AI agents into production. It is fast. It is also the reason most of our projects ship: a fixed date forces every decision that a pilot would postpone to be taken in the first week.",[483,487,488],{},"This is the playbook we follow. Every step ends with something written down and someone’s name next to it.",[490,491,493],"h2",{"id":492},"week-0-diagnose","Week 0: diagnose",[483,495,496],{},"It starts with a free call. If there is a fit, we spend the next week inside your operations, most of it with the people who do the work. We sit in on the inbox, the report, the supplier file. We want to see where the hours go and where the money leaks, not how the process is described in a slide.",[483,498,499],{},"The artefact is a map of your operations with the value of every agent we would build, and a fixed price. The sign-off is yours: which process goes first. We push for one with volume, clear rules and a number someone already cares about.",[490,501,503],{"id":502},"week-1-map","Week 1: map",[483,505,506],{},"This is the week most pilots skip, and the week that decides whether yours ships.",[508,509,510,518,524,530],"ol",{},[511,512,513,517],"li",{},[514,515,516],"strong",{},"Access."," Real credentials to the real systems: the CRM, the helpdesk, the ERP, the inbox. Not an export.",[511,519,520,523],{},[514,521,522],{},"Real cases."," A set of past emails, tickets or files, with what a good answer looked like. They become the tests.",[511,525,526,529],{},[514,527,528],{},"One metric, signed off."," First response time, time to spot a new listing, partner hours on admin. One number, owned by the team whose work changes, signed by the person who owns it.",[511,531,532,535],{},[514,533,534],{},"First approval rules."," What the agents may do alone, and what always goes to a person, written in plain language.",[483,537,538],{},"If week 1 ends without access and a signed metric, we say so, because it is cheaper for both sides than finding out on day 29.",[490,540,542],{"id":541},"week-2-build","Week 2: build",[483,544,545],{},"Now we build the agents, the rules, the shared memory and the command center, wired into your tools. The agents live where your team already works; nobody gets a new app to learn.",[483,547,548],{},"The real cases from week 1 become evals: automated tests that run every answer against what a good answer looked like. Customer-facing answers get a judge. Tone is agreed with the people who own it, often in a single working session with a pile of past replies.",[483,550,551],{},"The sign-off at the end of the week is a walkthrough with the team lead, on their own cases.",[490,553,555],{"id":554},"week-3-prove","Week 3: prove",[483,557,558],{},"First the evals, on real cases the agents have never seen. Then shadow mode on live work: the agents draft, people send. Every edit a person makes is a signal, and we read them daily.",[483,560,561],{},"This is also where the handoff rules are tuned with the team. Which cases go straight to a person, at what threshold, through which channel. The team should be able to change a rule themselves, and in week 3 they practise doing it.",[563,564,565,581],"table",{},[566,567,568],"thead",{},[569,570,571,575,578],"tr",{},[572,573,574],"th",{},"STEP",[572,576,577],{},"ARTEFACT",[572,579,580],{},"WHO SIGNS",[582,583,584,597,610,623,635],"tbody",{},[569,585,586,591,594],{},[587,588,589],"td",{},[514,590,70],{},[587,592,593],{},"Operations map, value per agent, fixed price",[587,595,596],{},"You: which process first",[569,598,599,604,607],{},[587,600,601],{},[514,602,603],{},"Week 1",[587,605,606],{},"Access, real cases, first approval rules",[587,608,609],{},"The owner of the metric",[569,611,612,617,620],{},[587,613,614],{},[514,615,616],{},"Week 2",[587,618,619],{},"Agents, memory, rules, command center, evals",[587,621,622],{},"The team lead, after a walkthrough",[569,624,625,629,632],{},[587,626,627],{},[514,628,78],{},[587,630,631],{},"Eval results, shadow-mode log, tuned handoffs",[587,633,634],{},"The team lead: ready to go live",[569,636,637,641,644],{},[587,638,639],{},[514,640,82],{},[587,642,643],{},"Production, monitoring, trained team",[587,645,646],{},"Both of us, against the metric",[490,648,650],{"id":649},"day-30-live","Day 30: live",[483,652,653],{},"On day 30 the agents are in production, on live work, monitored from the command center. The team is trained on it: how to read a run, how to approve, how to pause an agent, how to change a rule.",[483,655,656],{},"Production means the agents answer, route or update on their own inside the rules, and the number from week 1 is tracked every day from then on. It does not mean “available on request” or “ready for phase two”.",[490,658,660],{"id":659},"why-the-date-holds","Why the date holds",[483,662,663],{},"Three things keep the date honest. The scope is one process, not the company. The metric is agreed before a line of code, so nobody can move the goalposts later. And we put our money on it: live in 30 days, or your money back.",[483,665,666],{},"After day 30 we run the fleet: monitoring, tuning and, when you are ready, the next squad. On our largest project that became one new squad a month, each built on the memory and rules of the ones before it.",{"title":15,"searchDepth":16,"depth":16,"links":668},[669,670,671,672,673,674],{"id":492,"depth":16,"text":493},{"id":502,"depth":16,"text":503},{"id":541,"depth":16,"text":542},{"id":554,"depth":16,"text":555},{"id":649,"depth":16,"text":650},{"id":659,"depth":16,"text":660},"playbooks","2026-08-18","The 30-day playbook, week by ==week.==","30",{},"\u002Finsights\u002Fthe-30-day-playbook-week-by-week",4,[118,117,683],"build-buy-or-both",{"title":475,"description":485},"insights\u002Fthe-30-day-playbook-week-by-week","From the first call to production: the exact steps, artefacts and sign-offs we use.",[688,689,690,691,692,693],{"id":492,"label":493},{"id":502,"label":503},{"id":541,"label":542},{"id":554,"label":555},{"id":649,"label":650},{"id":659,"label":660},"rT6HBQ-x40ciVZH-qHpcDH1wLuGv79MQR2E3ffHF_dY",{"id":696,"title":697,"author":698,"blobHue":198,"body":699,"category":675,"date":920,"description":703,"draft":18,"extension":32,"featured":18,"headline":921,"hue":86,"letter":191,"meta":922,"navigation":89,"path":923,"readMinutes":681,"related":924,"seo":926,"stem":927,"summary":928,"toc":929,"__hash__":935},"insights_en\u002Finsights\u002Fwhat-agents-should-never-do-alone.md","What agents should never do alone",{"name":477,"role":478,"bio":479},{"type":12,"value":700,"toc":913},[701,704,707,711,717,723,729,732,736,739,742,745,749,752,773,776,780,783,891,894,900,904,907,910],[483,702,703],{},"The question we hear most from operations leads is not “can the agent do this?”. It is “what stops it doing something stupid?”. The honest answer is not a better prompt. It is a short list of rules, written by the people who own the risk, that the agent cannot talk its way around.",[483,705,706],{},"Writing those rules is less work than it sounds. Almost everything that should stay with a person falls into three kinds of decision.",[490,708,710],{"id":709},"three-kinds-of-decisions","Three kinds of decisions",[483,712,713,716],{},[514,714,715],{},"Money."," Refunds, credits, discounts, prices, payments. Anything that moves money in or out of the business, or changes what a customer pays.",[483,718,719,722],{},[514,720,721],{},"Promises."," A delivery date, a replacement, an exception to the policy, anything that commits the company to a customer or a supplier. A wrong answer is fixable. A wrong promise has to be honoured or broken.",[483,724,725,728],{},[514,726,727],{},"Anything you can’t undo."," Deleting records, sending to your whole customer base, cancelling or changing an order that is already moving. If the mistake cannot be reversed by clicking something, a person decides.",[483,730,731],{},"Everything else, like reading, classifying, drafting, looking up, summarising and routing, is usually safe for an agent to do alone, provided it is visible afterwards.",[490,733,735],{"id":734},"draft-then-apply","Draft, then apply",[483,737,738],{},"The most useful distinction in approval rules is between preparing an action and executing it. An agent can do all the work of a refund: find the order, check the policy, calculate the amount, write the reply. Then it stops, and a person applies it with one tap.",[483,740,741],{},"This keeps most of the time saving and almost none of the risk. The person is not doing the work. They are checking a finished proposal, with the order, the policy and the amount on one screen.",[483,743,744],{},"Over time, some drafts earn the right to apply themselves. When a person has approved the same kind of action enough times without changes, you can move it below the threshold. That is a decision for the team that owns it, not for the agent.",[490,746,748],{"id":747},"thresholds-not-feelings","Thresholds, not feelings",[483,750,751],{},"“Ask me when it’s important” is not a rule. An agent cannot know what feels important to you. A rule needs a number, a category or a list:",[753,754,755,761,767],"ul",{},[511,756,757,760],{},[514,758,759],{},"A number."," Refunds up to a limit are applied; above it, they go to the ops lead.",[511,762,763,766],{},[514,764,765],{},"A category."," Any change to a contract goes to legal, whatever the value.",[511,768,769,772],{},[514,770,771],{},"A list."," New suppliers, key accounts and anything to the press always go to a person.",[483,774,775],{},"Every rule names who decides and where they are asked: in Slack, Teams or email, wherever that person already works. A rule that sends approvals to a dashboard nobody opens is a rule that stops the business.",[490,777,779],{"id":778},"a-worked-example","A worked example",[483,781,782],{},"This is the shape of a rule set for an online store’s support and commerce agents. The limits are yours to set; the structure is what matters.",[563,784,785,798],{},[566,786,787],{},[569,788,789,792,795],{},[572,790,791],{},"ACTION",[572,793,794],{},"AGENT ALONE",[572,796,797],{},"GOES TO A PERSON",[582,799,800,813,826,839,852,865,878],{},[569,801,802,807,810],{},[587,803,804],{},[514,805,806],{},"Order status, tracking, policy questions",[587,808,809],{},"Answers",[587,811,812],{},"Never, unless the customer asks for one",[569,814,815,820,823],{},[587,816,817],{},[514,818,819],{},"Refund",[587,821,822],{},"Drafts every one, applies small ones",[587,824,825],{},"Above €500 in this example, the ops lead",[569,827,828,833,836],{},[587,829,830],{},[514,831,832],{},"Voucher or discount",[587,834,835],{},"Proposes, inside the margin floor",[587,837,838],{},"Anything outside the agreed limits",[569,840,841,846,849],{},[587,842,843],{},[514,844,845],{},"Price change",[587,847,848],{},"Never writes a price",[587,850,851],{},"The pricing owner, after the margin check",[569,853,854,859,862],{},[587,855,856],{},[514,857,858],{},"Delivery date",[587,860,861],{},"Quotes what the courier data says",[587,863,864],{},"Any exception or guarantee",[569,866,867,872,875],{},[587,868,869],{},[514,870,871],{},"Order change or cancellation",[587,873,874],{},"Drafts",[587,876,877],{},"Always applied by a person",[569,879,880,885,888],{},[587,881,882],{},[514,883,884],{},"Campaign to the whole base",[587,886,887],{},"Prepares and simulates",[587,889,890],{},"Always sent by a person",[483,892,893],{},"Start with the right-hand column. The right-hand column is the list of things your team has decided to keep. Everything to its left is work they no longer have to do.",[895,896,897],"blockquote",{},[483,898,899],{},"If you can’t undo it, a person does it.",[490,901,903],{"id":902},"the-veto-that-isnt-a-model","The veto that isn’t a model",[483,905,906],{},"Some rules are too important to leave to a language model, even one that is being checked. Margin is the clearest case.",[483,908,909],{},"In the marketplace system we built, no agent ever writes a price. Agents can propose a discount or a clearance offer, but every proposal passes through a margin guardrail first. The guardrail is plain code, not a model. It calculates the floor for that product after returns, fees and shipping, and anything below it is vetoed. There is nothing to persuade and no prompt to get wrong.",[483,911,912],{},"That is the pattern we use wherever a rule can be expressed as arithmetic or a lookup. The model proposes, deterministic code checks, and a person approves what the rules say a person approves. Each layer does the thing it is good at, and none of them is asked to be the last line of defence alone.",{"title":15,"searchDepth":16,"depth":16,"links":914},[915,916,917,918,919],{"id":709,"depth":16,"text":710},{"id":734,"depth":16,"text":735},{"id":747,"depth":16,"text":748},{"id":778,"depth":16,"text":779},{"id":902,"depth":16,"text":903},"2026-09-08","What agents should never do ==alone.==",{},"\u002Finsights\u002Fwhat-agents-should-never-do-alone",[452,116,925],"what-an-agentic-os-is-and-what-it-isnt",{"title":697,"description":703},"insights\u002Fwhat-agents-should-never-do-alone","A practical way to write approval rules: money, promises and anything you can’t undo.",[930,931,932,933,934],{"id":709,"label":710},{"id":734,"label":735},{"id":747,"label":748},{"id":778,"label":779},{"id":902,"label":903},"mWnCbu3ZKOI9CJrpjxbakDDnQwkJuenNM8EI7HSjdMA",{"id":937,"title":938,"author":939,"blobHue":105,"body":940,"category":1127,"date":1128,"description":1129,"draft":18,"extension":32,"featured":89,"headline":1130,"hue":97,"letter":1131,"meta":1132,"navigation":89,"path":1133,"readMinutes":1134,"related":1135,"seo":1136,"stem":1137,"summary":1138,"toc":1139,"__hash__":1145},"insights_en\u002Finsights\u002Fwhy-95-percent-of-ai-pilots-never-reach-production.md","Why 95% of AI pilots never reach production",{"name":477,"role":478,"bio":479},{"type":12,"value":941,"toc":1120},[942,950,953,957,962,967,971,974,994,998,1001,1082,1086,1107,1111,1114],[483,943,944,945,949],{},"In 2025, MIT’s NANDA initiative looked at how companies were using generative AI and found that about 95% of enterprise pilots delivered no measurable impact on the P&L.",[946,947,948],"sup",{},"1"," Billions spent, demos applauded, and almost nothing changed in how the business actually ran.",[483,951,952],{},"We’ve seen the same pattern from the inside. Most of the companies that call us have already run a pilot. It worked in the demo. It never went live.",[490,954,956],{"id":955},"the-number-everyone-quotes","The number everyone quotes",[483,958,959,960],{},"The headline is easy to misread. It doesn’t say AI doesn’t work. It says pilots don’t turn into production. The same report found that projects built with specialised external partners reached deployment about twice as often as internal builds, and that the biggest returns came from unglamorous back-office work, not from customer-facing chatbots.",[946,961,948],{},[895,963,964],{},[483,965,966],{},"The model is rarely the problem. The plumbing is.",[490,968,970],{"id":969},"its-not-the-model","It’s not the model",[483,972,973],{},"When we look at why a pilot stalled, the answer is almost never “the AI wasn’t smart enough”. It is one of three things:",[508,975,976,982,988],{},[511,977,978,981],{},[514,979,980],{},"No live data."," The pilot ran on an export. Connecting it to the real ERP, inbox or helpdesk was “phase two”, and phase two never came.",[511,983,984,987],{},[514,985,986],{},"No owner."," The innovation team built it. The team whose work it would take never asked for it, and never adopted it.",[511,989,990,993],{},[514,991,992],{},"No number."," Success was “a good demo”. Nobody agreed what had to move, so nobody could say it had worked.",[490,995,997],{"id":996},"four-things-the-5-do","Four things the 5% do",[483,999,1000],{},"The projects that ship look different from day one. They connect to live systems in week one, not month six. The metric is agreed before a line of code is written, and it belongs to the team whose work changes. People approve the decisions that carry risk, so nobody has to trust the agent blindly. And there is a date: production in weeks, not a roadmap.",[563,1002,1003,1015],{},[566,1004,1005],{},[569,1006,1007,1009,1012],{},[572,1008],{},[572,1010,1011],{},"A PILOT",[572,1013,1014],{},"A SYSTEM IN PRODUCTION",[582,1016,1017,1030,1043,1056,1069],{},[569,1018,1019,1024,1027],{},[587,1020,1021],{},[514,1022,1023],{},"Data",[587,1025,1026],{},"A sample export",[587,1028,1029],{},"Your live systems",[569,1031,1032,1037,1040],{},[587,1033,1034],{},[514,1035,1036],{},"Success",[587,1038,1039],{},"A good demo",[587,1041,1042],{},"A number agreed on day one",[569,1044,1045,1050,1053],{},[587,1046,1047],{},[514,1048,1049],{},"Owner",[587,1051,1052],{},"The innovation team",[587,1054,1055],{},"The team whose work it takes",[569,1057,1058,1063,1066],{},[587,1059,1060],{},[514,1061,1062],{},"Humans",[587,1064,1065],{},"Watching",[587,1067,1068],{},"Approving what matters",[569,1070,1071,1076,1079],{},[587,1072,1073],{},[514,1074,1075],{},"Timeline",[587,1077,1078],{},"Open-ended",[587,1080,1081],{},"Weeks",[490,1083,1085],{"id":1084},"a-checklist-before-you-start","A checklist before you start",[1087,1088,1090],"prose-checklist",{"title":1089},"Before you approve the next AI project, ask",[753,1091,1092,1095,1098,1101,1104],{},[511,1093,1094],{},"Which live system does it read and write, in week one?",[511,1096,1097],{},"Which number moves, and who owns it?",[511,1099,1100],{},"Which decisions stay with a person?",[511,1102,1103],{},"What happens on day 30?",[511,1105,1106],{},"Who owns the code, the prompts and the data?",[490,1108,1110],{"id":1109},"what-this-means-for-you","What this means for you",[483,1112,1113],{},"If your last pilot is gathering dust, it probably wasn’t a bad idea. It was missing the boring parts. Start from one process that hurts, connect it to the real systems, agree the number and put a date on it. That’s the whole method. It isn’t magic; it’s engineering.",[1115,1116,1117],"prose-footnotes",{},[483,1118,1119],{},"¹ MIT NANDA, “The GenAI Divide: State of AI in Business 2025”.",{"title":15,"searchDepth":16,"depth":16,"links":1121},[1122,1123,1124,1125,1126],{"id":955,"depth":16,"text":956},{"id":969,"depth":16,"text":970},{"id":996,"depth":16,"text":997},{"id":1084,"depth":16,"text":1085},{"id":1109,"depth":16,"text":1110},"strategy","2026-09-22","In 2025, MIT’s NANDA initiative looked at how companies were using generative AI and found that about 95% of enterprise pilots delivered no measurable impact on the P&L.1 Billions spent, demos applauded, and almost nothing changed in how the business actually ran.","Why 95% of AI pilots never reach ==production.==","95",{},"\u002Finsights\u002Fwhy-95-percent-of-ai-pilots-never-reach-production",7,[925,452,116],{"title":938,"description":1129},"insights\u002Fwhy-95-percent-of-ai-pilots-never-reach-production","It’s rarely the model. It’s the data, the owner and the missing number. Here’s what the ones that ship do differently.",[1140,1141,1142,1143,1144],{"id":955,"label":956},{"id":969,"label":970},{"id":996,"label":997},{"id":1084,"label":1085},{"id":1109,"label":1110},"9wJDwNO8uXgEwGK-cE1WYthNEN-J5Ic1e7kI_crM30g",1790881957480]