[{"data":1,"prerenderedAt":975},["ShallowReactive",2],{"locale-routes":3,"squeeze-paths":247,"announcements":248,"page-\u002Fen\u002Finsights\u002Fai-software-development-agency-size":249,"insights-author-\u002Fen\u002Finsights\u002Fai-software-development-agency-size":780},{"home":4,"shopware-margin-check":7,"shopware-plugin-report":10,"privacy":13,"framework-focus":16,"framework":19,"framework-principles":22,"imprint":25,"contact":28,"newsletter-confirmed":31,"shopware-agency-b2b":34,"shopware-agency-b2b-configurator":37,"shopware-agency-electronics":40,"shopware-agency-development":43,"shopware-extension-b2b-quote-management":46,"shopware-agency-geo":49,"shopware-agency":52,"shopware-agency-pricing":55,"shopware-cost-calculator":58,"shopware-agency-migration":61,"shopware-agency-seo":64,"shopware-agency-support":67,"shopware-agency-switch":70,"location-hamburg":73,"location-munich":76,"about":79,"about-mission":82,"about-partners":85,"services-consulting":88,"services-development":91,"services":94,"services-optimization":97,"services-growth":100,"insights-ai-sales-channels-shopware":103,"insights-b2b-quote-management-release":106,"insights-b2b-shop-starten-checkliste":109,"insights-comspot-trade-in":112,"insights-ecommerce-deadlines-2026-27":115,"insights-e-invoice-b2b-shopware-xrechnung-zugferd":118,"insights-ecommerce-search-that-finds":121,"insights-warranty-notice-vs-garan-label":124,"insights-warranty-label-modal":127,"insights-warranty-label-shopware-6":130,"insights-handelskoenig-b2b-quotes":133,"insights-ai-agency-size":136,"insights-kpi-definitions-ecommerce":139,"insights-measurable-ecommerce-strategy":142,"insights-php2shell-shopware":145,"insights-shopware-5-eol":148,"insights-shopware-6-7-release":151,"insights-shopware-6-8-outlook":154,"insights-shopware-6-plans-pricing-2026":157,"insights-shopware-6-migration-noetig":160,"insights-shopware-gmv-2026":163,"insights-shopware-vs-shopify-2026":166,"insights-shopware-maintenance-retainer-development":169,"insights-withdrawal-button-shopware-6":172,"whitepaper-agentic-commerce-b2b-beschaffung":175,"whitepaper-agenturwechsel-wartungsprojekt":178,"whitepaper-compliance-2026-27":181,"whitepapers":184,"whitepaper-kanalverbund-marktplatz-shop-agentic":187,"whitepaper-shopware-6-boutique-vs-grossagentur-tco":190,"case-comspot-focus":193,"case-converge-focus":196,"case-framily-focus":199,"case-ghs-waescherutsche-focus":202,"case-handelskoenig-focus":205,"case-lupa-focus":208,"case-shifter-focus":211,"case-searchhub-comspot":214,"case-topi-focus":217,"case-xucker-focus":220,"talks-beyond-seo-scuc-2026":223,"talks":226,"talks-shopware-container-hosting-hitc-2026":229,"talks-symfony-cli-docker-scuc-2025":232,"webinars":235,"team-alex-jank":238,"team-benedikt-rillox":241,"team":244},{"de":5,"en":6},"\u002F","\u002Fen",{"de":8,"en":9},"\u002Fshopware-margen-check","\u002Fen\u002Fshopware-margin-check",{"de":11,"en":12},"\u002Fshopware-plugin-report","\u002Fen\u002Fshopware-plugin-report",{"de":14,"en":15},"\u002Fdatenschutz","\u002Fen\u002Fprivacy",{"de":17,"en":18},"\u002Fframework\u002Ffocus","\u002Fen\u002Fframework\u002Ffocus",{"de":20,"en":21},"\u002Fframework","\u002Fen\u002Fframework",{"de":23,"en":24},"\u002Fframework\u002Fprinzipien","\u002Fen\u002Fframework\u002Fprinciples",{"de":26,"en":27},"\u002Fimpressum","\u002Fen\u002Fimprint",{"de":29,"en":30},"\u002Fkontakt","\u002Fen\u002Fcontact",{"de":32,"en":33},"\u002Fnewsletter\u002Fbestaetigt","\u002Fen\u002Fnewsletter\u002Fconfirmed",{"de":35,"en":36},"\u002Fshopware-agentur\u002Fb2b","\u002Fen\u002Fshopware-agency\u002Fb2b",{"de":38,"en":39},"\u002Fshopware-agentur\u002Fb2b-konfigurator","\u002Fen\u002Fshopware-agency\u002Fb2b-configurator",{"de":41,"en":42},"\u002Fshopware-agentur\u002Felektronikhandel","\u002Fen\u002Fshopware-agency\u002Felectronics-retail",{"de":44,"en":45},"\u002Fshopware-agentur\u002Fentwicklung","\u002Fen\u002Fshopware-agency\u002Fdevelopment",{"de":47,"en":48},"\u002Fshopware-agentur\u002Fentwicklung\u002Fb2b-angebotsverwaltung-preisanfragen","\u002Fen\u002Fshopware-agency\u002Fdevelopment\u002Fb2b-quote-management-price-requests",{"de":50,"en":51},"\u002Fshopware-agentur\u002Fgeo","\u002Fen\u002Fshopware-agency\u002Fgeo",{"de":53,"en":54},"\u002Fshopware-agentur","\u002Fen\u002Fshopware-agency",{"de":56,"en":57},"\u002Fshopware-agentur\u002Fkosten","\u002Fen\u002Fshopware-agency\u002Fpricing",{"de":59,"en":60},"\u002Fshopware-agentur\u002Fkosten-rechner","\u002Fen\u002Fshopware-agency\u002Fcost-calculator",{"de":62,"en":63},"\u002Fshopware-agentur\u002Fmigration","\u002Fen\u002Fshopware-agency\u002Fmigration",{"de":65,"en":66},"\u002Fshopware-agentur\u002Fseo","\u002Fen\u002Fshopware-agency\u002Fseo",{"de":68,"en":69},"\u002Fshopware-agentur\u002Fsupport","\u002Fen\u002Fshopware-agency\u002Fsupport",{"de":71,"en":72},"\u002Fshopware-agentur\u002Fwechsel","\u002Fen\u002Fshopware-agency\u002Fswitch",{"de":74,"en":75},"\u002Fstandorte\u002Fhamburg","\u002Fen\u002Flocations\u002Fhamburg",{"de":77,"en":78},"\u002Fstandorte\u002Fmuenchen","\u002Fen\u002Flocations\u002Fmunich",{"de":80,"en":81},"\u002Fueber-uns","\u002Fen\u002Fabout",{"de":83,"en":84},"\u002Fueber-uns\u002Fmission","\u002Fen\u002Fabout\u002Fmission",{"de":86,"en":87},"\u002Fueber-uns\u002Fpartner","\u002Fen\u002Fabout\u002Fpartners",{"de":89,"en":90},"\u002Fleistungen\u002Fberatung","\u002Fen\u002Fservices\u002Fconsulting",{"de":92,"en":93},"\u002Fleistungen\u002Fentwicklung","\u002Fen\u002Fservices\u002Fdevelopment",{"de":95,"en":96},"\u002Fleistungen","\u002Fen\u002Fservices",{"de":98,"en":99},"\u002Fleistungen\u002Foptimierung","\u002Fen\u002Fservices\u002Foptimization",{"de":101,"en":102},"\u002Fleistungen\u002Fwachstum","\u002Fen\u002Fservices\u002Fgrowth",{"de":104,"en":105},"\u002Finsights\u002Fai-verkaufskanaele-shopware","\u002Fen\u002Finsights\u002Fai-sales-channels-shopware",{"de":107,"en":108},"\u002Finsights\u002Fb2b-angebotsverwaltung-shopware-store-release","\u002Fen\u002Finsights\u002Fb2b-quote-management-shopware-store-release",{"de":110,"en":111},"\u002Finsights\u002Fb2b-shop-starten-checkliste-shopware","\u002Fen\u002Finsights\u002Fstart-b2b-store-shopware-checklist",{"de":113,"en":114},"\u002Finsights\u002Fcomspot-trade-in-altgeraete","\u002Fen\u002Finsights\u002Fcomspot-trade-in-device-buyback",{"de":116,"en":117},"\u002Finsights\u002Fe-commerce-fristen-2026-27","\u002Fen\u002Finsights\u002Fe-commerce-deadlines-2026-27",{"de":119,"en":120},"\u002Finsights\u002Fe-rechnung-b2b-shopware-xrechnung-zugferd","\u002Fen\u002Finsights\u002Fe-invoice-b2b-shopware-xrechnung-zugferd",{"de":122,"en":123},"\u002Finsights\u002Fecommerce-suche-die-findet","\u002Fen\u002Finsights\u002Fecommerce-search-that-finds",{"de":125,"en":126},"\u002Finsights\u002Fgewaehrleistungshinweis-vs-garan-label","\u002Fen\u002Finsights\u002Fwarranty-notice-vs-garan-label",{"de":128,"en":129},"\u002Finsights\u002Fgewaehrleistungslabel-modal-zulaessig","\u002Fen\u002Finsights\u002Fwarranty-label-modal-allowed",{"de":131,"en":132},"\u002Finsights\u002Fgewaehrleistungslabel-shopware-6","\u002Fen\u002Finsights\u002Fwarranty-label-shopware-6",{"de":134,"en":135},"\u002Finsights\u002Fhandelskoenig-b2b-angebotsanfragen","\u002Fen\u002Finsights\u002Fhandelskoenig-b2b-quote-requests",{"de":137,"en":138},"\u002Finsights\u002Fki-softwareentwicklung-agenturgroesse","\u002Fen\u002Finsights\u002Fai-software-development-agency-size",{"de":140,"en":141},"\u002Finsights\u002Fkpi-definitionen-e-commerce","\u002Fen\u002Finsights\u002Fkpi-definitions-ecommerce",{"de":143,"en":144},"\u002Finsights\u002Fmessbare-e-commerce-strategie","\u002Fen\u002Finsights\u002Fmeasurable-ecommerce-strategy",{"de":146,"en":147},"\u002Finsights\u002Fphp-sicherheitsluecke-php2shell-shopware","\u002Fen\u002Finsights\u002Fphp-vulnerability-php2shell-shopware",{"de":149,"en":150},"\u002Finsights\u002Fshopware-5-eol","\u002Fen\u002Finsights\u002Fshopware-5-end-of-life",{"de":152,"en":153},"\u002Finsights\u002Fshopware-6-7-release-update-vorteile","\u002Fen\u002Finsights\u002Fshopware-6-7-release-update-benefits",{"de":155,"en":156},"\u002Finsights\u002Fshopware-6-8-ausblick-roadmap-vorbereitung","\u002Fen\u002Finsights\u002Fshopware-6-8-outlook-roadmap-preparation",{"de":158,"en":159},"\u002Finsights\u002Fshopware-6-lizenzen-kosten-2026","\u002Fen\u002Finsights\u002Fshopware-6-plans-pricing-2026",{"de":161,"en":162},"\u002Finsights\u002Fshopware-6-migration-noetig","\u002Fen\u002Finsights\u002Fshopware-6-migration-necessary",{"de":164,"en":165},"\u002Finsights\u002Fshopware-gmv-berechnung-1-million-grenze","\u002Fen\u002Finsights\u002Fshopware-gmv-calculation-one-million-threshold",{"de":167,"en":168},"\u002Finsights\u002Fshopware-vs-shopify-vergleich-2026","\u002Fen\u002Finsights\u002Fshopware-vs-shopify-comparison-2026",{"de":170,"en":171},"\u002Finsights\u002Fshopware-wartung-retainer-weiterentwicklung","\u002Fen\u002Finsights\u002Fshopware-maintenance-retainer-development",{"de":173,"en":174},"\u002Finsights\u002Fwiderrufsbutton-shopware-6","\u002Fen\u002Finsights\u002Fwithdrawal-button-shopware-6",{"de":176,"en":177},"\u002Fwhitepapers\u002Fagentic-commerce-b2b-beschaffung","\u002Fen\u002Fwhitepapers\u002Fagentic-commerce-b2b-procurement",{"de":179,"en":180},"\u002Fwhitepapers\u002Fagenturwechsel-wartungsprojekt","\u002Fen\u002Fwhitepapers\u002Fagency-switch-maintenance-project",{"de":182,"en":183},"\u002Fwhitepapers\u002Fe-commerce-compliance-2026-27","\u002Fen\u002Fwhitepapers\u002Fe-commerce-compliance-2026-27",{"de":185,"en":186},"\u002Fwhitepapers","\u002Fen\u002Fwhitepapers",{"de":188,"en":189},"\u002Fwhitepapers\u002Fkanalverbund-marktplatz-shop-agentic","\u002Fen\u002Fwhitepapers\u002Fchannel-network-marketplace-shop-agentic",{"de":191,"en":192},"\u002Fwhitepapers\u002Fshopware-6-boutique-vs-grossagentur-tco","\u002Fen\u002Fwhitepapers\u002Fshopware-6-boutique-vs-large-agency-tco",{"de":194,"en":195},"\u002Fcase-studies\u002Fcomspot-focus","\u002Fen\u002Fcase-studies\u002Fcomspot-focus",{"de":197,"en":198},"\u002Fcase-studies\u002Fconverge-focus","\u002Fen\u002Fcase-studies\u002Fconverge-focus",{"de":200,"en":201},"\u002Fcase-studies\u002Fframily-focus","\u002Fen\u002Fcase-studies\u002Fframily-focus",{"de":203,"en":204},"\u002Fcase-studies\u002Fghs-waescherutsche-focus","\u002Fen\u002Fcase-studies\u002Fghs-waescherutsche-focus",{"de":206,"en":207},"\u002Fcase-studies\u002Fhandelskoenig-focus","\u002Fen\u002Fcase-studies\u002Fhandelskoenig-focus",{"de":209,"en":210},"\u002Fcase-studies\u002Flupa-focus","\u002Fen\u002Fcase-studies\u002Flupa-focus",{"de":212,"en":213},"\u002Fcase-studies\u002Fshifter-focus","\u002Fen\u002Fcase-studies\u002Fshifter-focus",{"de":215,"en":216},"\u002Fcase-studies\u002Fshopware-suchintegration","\u002Fen\u002Fcase-studies\u002Fshopware-search-integration",{"de":218,"en":219},"\u002Fcase-studies\u002Ftopi-focus","\u002Fen\u002Fcase-studies\u002Ftopi-focus",{"de":221,"en":222},"\u002Fcase-studies\u002Fxucker-focus","\u002Fen\u002Fcase-studies\u002Fxucker-focus",{"de":224,"en":225},"\u002Fvortraege\u002Fbeyond-seo-scuc-2026","\u002Fen\u002Ftalks\u002Fbeyond-seo-scuc-2026",{"de":227,"en":228},"\u002Fvortraege","\u002Fen\u002Ftalks",{"de":230,"en":231},"\u002Fvortraege\u002Fshopware-container-hosting-hitc-2026","\u002Fen\u002Ftalks\u002Fshopware-container-hosting-hitc-2026",{"de":233,"en":234},"\u002Fvortraege\u002Fsymfony-cli-docker-scuc-2025","\u002Fen\u002Ftalks\u002Fsymfony-cli-docker-scuc-2025",{"de":236,"en":237},"\u002Fwebinare","\u002Fen\u002Fwebinars",{"de":239,"en":240},"\u002Fteam\u002Falex-jank","\u002Fen\u002Fteam\u002Falex-jank",{"de":242,"en":243},"\u002Fteam\u002Fbenedikt-rillox","\u002Fen\u002Fteam\u002Fbenedikt-rillox",{"de":245,"en":246},"\u002Fteam","\u002Fen\u002Fteam",[8,11,9,12],[],{"id":250,"title":251,"author":252,"body":253,"citations":749,"description":757,"draft":758,"extension":759,"image":760,"meta":763,"navigation":764,"ogImage":765,"path":138,"publishedAt":766,"readingTime":767,"robots":765,"schemaOrg":765,"seo":768,"sitemap":771,"squeeze":758,"stem":772,"tags":773,"translationKey":778,"updatedAt":765,"__hash__":779},"insights_en\u002Fen\u002Finsights\u002Fai-software-development-agency-size.md","AI in software development: why agency size is no longer an argument","benedikt-rillox",{"type":254,"value":255,"toc":725},"minimark",[256,260,263,266,271,287,290,295,307,310,313,317,330,333,342,346,349,352,355,359,367,370,373,376,380,383,387,390,399,403,406,409,413,416,427,431,434,437,441,444,450,456,462,471,474,483,487,490,542,545,549,552,562,571,577,580,584,591,594,601,605,608,628,643,647,694,698],[257,258,259],"p",{},"There is one line that has worked in agency pitches for twenty years: \"We\nhave 120 employees.\" The number itself is never the point. The point is a\npromise — resilience, speed, scalability. If someone falls ill, the next\nperson takes over. If things get tight, we put two more people on it.",[257,261,262],{},"That promise came from a world in which output was tied fairly directly to\nperson-hours. Anyone who wanted to deliver twice as much needed roughly twice\nas many hands. In that world, size genuinely was an argument.",[257,264,265],{},"That world is gone — at least at the point where it used to make the\ndifference. AI has changed the pace of software development less than the\nmarketing slides claim and more than the sceptics admit. Above all, it has\nmoved the bottleneck: away from writing code, towards judging code. And that\nshift affects large and small service providers in completely different ways.",[267,268,270],"h2",{"id":269},"what-the-data-says-and-what-it-doesnt","What the data says — and what it doesn't",[257,272,273,274,278,279,286],{},"Let's start with the uncomfortable number. In 2025, METR ran a randomised\ncontrolled trial with 16 experienced open-source developers across 246 real\nissues in their own, familiar repositories. The expectation was a 24 percent\nspeedup. What was measured was the opposite: with AI tools, the developers\ntook ",[275,276,277],"strong",{},"19 percent longer",". More remarkable still is the self-assessment\nafterwards — the participants believed they had been 20 percent faster\n(",[280,281,285],"a",{"href":282,"rel":283},"https:\u002F\u002Fmetr.org\u002Fblog\u002F2025-07-10-early-2025-ai-experienced-os-dev-study\u002F",[284],"nofollow","METR",").",[257,288,289],{},"That gap between perception and measurement is the most important finding of\nthe past two years. It explains why almost every developer reports massive\nproductivity gains while project timelines in many companies have not moved.\nAI feels fast because the phase of staring at an empty file disappears. What\ncomes after it — reading, checking, discarding, adjusting — does not feel like\nwork on the task, but that is exactly what it is.",[291,292,294],"h3",{"id":293},"twelve-months-later-the-picture-changes","Twelve months later the picture changes",[257,296,297,298,301,302,286],{},"Anyone quoting those 19 percent today as evidence against AI is quoting stale\ndata — METR itself now flags the study as out of date. In February 2026 the\nteam published a continuation with 57 developers, 143 repositories and more\nthan 800 tasks. For the ten developers who also took part in the first round,\nthe sign flips: an estimated ",[275,299,300],{},"18 percent faster",", with a confidence interval\nrunning from 38 percent faster to 9 percent slower. Among newly recruited\nparticipants the effect stays in the noise at 4 percent\n(",[280,303,306],{"href":304,"rel":305},"https:\u002F\u002Fmetr.org\u002Fblog\u002F2026-02-24-uplift-update\u002F",[284],"METR, February 2026",[257,308,309],{},"That comes with the caveat METR itself emphasises: the study now suffers from\nselection effects. Developers decline to participate because they do not want\nto work without AI, and between 30 and 50 percent reported deliberately not\nsubmitting tasks — precisely the tasks where AI would have had the biggest\nleverage. The real effect may therefore be larger than the measured one.",[257,311,312],{},"Two things hold up regardless. First: within twelve months the tools moved\nfrom \"slows experienced developers down\" to \"helps those who have mastered\nthem\". Second: the gain sits with the person, not the tool. Same models, same\ntasks — and the difference between regression and progress came down to who\noperates them and how much practice they have.",[291,314,316],{"id":315},"the-verification-tax","The verification tax",[257,318,319,320,323,324,329],{},"Google Cloud's DORA research provides the wide-angle picture: 90 percent of\ntechnology professionals now use AI at work, and over 80 percent believe it\nhas made them more productive. At the same time, higher AI adoption correlates\nnot only with higher throughput but also with higher ",[275,321,322],{},"delivery instability","\n(",[280,325,328],{"href":326,"rel":327},"https:\u002F\u002Fdora.dev\u002Finsights\u002Fbalancing-ai-tensions\u002F",[284],"DORA","). More changes per\nweek, more of them going wrong.",[257,331,332],{},"DORA calls the mechanism behind this the verification tax: the time saved\nwriting moves into reviewing. Review turns from a quality ritual into the\nbottleneck, and the cognitive load shifts from whoever produces to whoever\nreads it back. Teams without a solid verification system pay that tax not in\ntime but in production incidents.",[257,334,335,336,341],{},"The Stack Overflow Developer Survey 2026 shows that developers themselves work\nexactly that way: 48 percent trust AI output only when they can easily verify\nit. A full 6.6 percent would trust it for important decisions\n(",[280,337,340],{"href":338,"rel":339},"https:\u002F\u002Fsurvey.stackoverflow.co\u002F2026\u002Fai\u002Fdata\u002Fai-trust",[284],"Stack Overflow 2026",").\nThat is not technophobia, it is a working method: let it generate, then decide\nyourself.",[267,343,345],{"id":344},"ai-is-an-amplifier-not-a-leveller","AI is an amplifier, not a leveller",[257,347,348],{},"The sentence worth keeping from DORA's work is this: in software development,\nAI is primarily an amplifier. It magnifies the strengths of high-performing\norganisations and the dysfunctions of struggling ones.",[257,350,351],{},"That is the opposite of what many expected. The expectation was levelling — if\nthe model writes the code, it no longer matters who is sitting in front of it.\nWhat actually happens is the reverse. Where clean architecture, clear\ninterfaces, real test coverage and a functioning review process exist, the\namplifier turns into speed. Where they are missing, it turns into technical\ndebt — at the same pace.",[257,353,354],{},"And this is where the actual subject of this article begins: none of what\nmakes the amplifier usable is a function of headcount. Clean domain modelling\ndoes not emerge from more people working on it. A good stack decision does not\nimprove because a committee makes it. Test coverage is a question of\ndiscipline, not capacity.",[267,356,358],{"id":357},"why-headcount-was-never-speed","Why headcount was never speed",[257,360,361,362,366],{},"In 1975, Fred Brooks described in ",[363,364,365],"em",{},"The Mythical Man-Month"," why adding\ndevelopers makes a late project later: communication overhead grows not\nlinearly with team size but quadratically. Six people have 15 possible\ncoordination pairs, twenty people have 190. Every one of those connections is\na place where context gets lost.",[257,368,369],{},"In agencies, a second layer is added that has nothing to do with communication\nand everything to do with the business model. The larger the organisation, the\nmore likely it is that whoever sells the project is not whoever builds it. In\nbetween sit handovers: from sales to consulting, from consulting to concept,\nfrom concept to development. At every handover a document gets written instead\nof a decision getting made.",[257,371,372],{},"Those handover costs were always there. They used to be offset by the fact\nthat real capacity stood on the other side: ten developers simply wrote more\nlines in a week than two. That offset is exactly what disappears. The leverage\nof raw writing capacity has fallen, the handover costs have stayed. What\nremains is a structural disadvantage.",[257,374,375],{},"This is not an argument against large service providers as such. For\nmulti-country rollouts with parallel teams, 24\u002F7 operating models, or projects\nin which ten specialist disciplines have to work simultaneously, size remains\nthe right answer. It is an argument against reading headcount as a promise of\nquality or speed. For the typical commerce project — one shop, one ERP, a\nhandful of interfaces, clear processes — it no longer says anything.",[267,377,379],{"id":378},"the-four-decisions-that-make-the-difference-now","The four decisions that make the difference now",[257,381,382],{},"If writing capacity is no longer the bottleneck, then the bottleneck sits with\nthe decisions. Four of them, specifically. Each one is either made by an\nexperienced person in hours or missed by an inexperienced one over weeks — and\nfrom the outside the two are indistinguishable at first, because in both cases\ncode appears and the code works.",[291,384,386],{"id":385},"_1-standard-or-custom-code","1. Standard or custom code",[257,388,389],{},"AI has dramatically lowered the cost of the first version of custom\ndevelopment. That is the most dangerous change of all, because it appears to\nshift the build-or-buy trade-off towards \"build\". Appears — because generated\ncode is an asset with running costs. It has to be read, updated, migrated and\ndragged along through every major release of the surrounding system.",[257,391,392,393,398],{},"In Shopware terms: if a requirement can be solved through the official\nextension points, that is where it is solved update-safely. If a model\nproduces a working core modification in twenty minutes, the requirement is met\ntoday and becomes a problem at the next major release. Shopware's own plugin\nguide names exactly the properties that make the difference later — idempotent\nmigrations, minimal lifecycle logic, encapsulated domain logic\n(",[280,394,397],{"href":395,"rel":396},"https:\u002F\u002Fdeveloper.shopware.com\u002Fdocs\u002Fguides\u002Fplugins\u002Fplugins\u002Fplugin-base-guide.html",[284],"Shopware Developer Docs",").\nBeing able to make that distinction is experience. The model does not make it;\nit implements what it is told.",[291,400,402],{"id":401},"_2-where-the-logic-lives","2. Where the logic lives",[257,404,405],{},"Pricing logic in the theme or in domain code. Discount rules in the ERP or in\nthe shop. Catalog transformation in the pipeline or in the frontend. These\ndecisions determine whether a system is extensible in two years or whether\nevery change touches three places.",[257,407,408],{},"This is where the amplifier shows most clearly. An AI agent let loose on a\nclean domain boundary delivers surprisingly consistent results, because the\nstructure constrains the solution space. The same agent in a system without\nclear boundaries produces plausible code in the wrong place — a great deal of\nit, very quickly.",[291,410,412],{"id":411},"_3-the-verification-system","3. The verification system",[257,414,415],{},"If the verification tax is real, then the verification system is the actual\nproductivity factor. That means automated tests that protect behavior rather\nthan implementation, a CI that genuinely blocks when it is red, staging\nenvironments that match production, monitoring from day one — and a human\nreview that does not wave things through.",[257,417,418,419,422,423,426],{},"This is where most AI expectations fail. A team that produces three times as\nmany changes while keeping the same review bottleneck has not gained\nthroughput, it has built a queue. So every discussion about AI speed has to\ninclude the question of who reads it back and against which standard. How we\norganise that is described in ",[280,420,421],{"href":93},"Development","; what\noperations look like afterwards is in\n",[280,424,425],{"href":171},"maintenance and continuous development",".",[291,428,430],{"id":429},"_4-what-does-not-get-built","4. What does not get built",[257,432,433],{},"The most effective decision is still the deletion. The cheaper implementation\nbecomes, the weaker the natural brake that effort used to provide. Features\nthat previously died on budget now get built — and every one of them has to be\nmaintained permanently.",[257,435,436],{},"An agency that says no here saves more money than any efficiency gain in\nimplementation. That does, however, require a business model that does not\ndepend on filling hours. Anyone who has to keep capacity on the bench has a\nstructural interest in using it.",[267,438,440],{"id":439},"the-new-risk-profile","The new risk profile",[257,442,443],{},"AI has not reduced project risk, it has relocated it. Four risks are larger\ntoday than they were two years ago:",[257,445,446,449],{},[275,447,448],{},"Plausible code that does not hold."," Generated code almost always looks\nright. It names variables sensibly, has a structure and produces no syntax\nerrors. The errors sit in edge cases, side effects and assumptions — exactly\nwhere a superficial code review does not look.",[257,451,452,455],{},[275,453,454],{},"Accelerated technical debt."," The DORA data shows the link between AI\nadoption and delivery instability. A system that receives twice as much code\nin six months as planned is not twice as far along; it is twice as hard to\nreason about.",[257,457,458,461],{},[275,459,460],{},"Loss of handover readiness."," Code nobody on the team can explain is not an\nasset, it is a dependency. The question \"could another team take this over?\"\nis the most honest quality check there is — and it becomes more important with\ngenerated code, not less.",[257,463,464,467,468,426],{},[275,465,466],{},"Security."," More code, faster cycles and dependencies added in passing all\nenlarge the attack surface. How quickly that turns into an acute incident is\nsomething we described in the case of the\n",[280,469,470],{"href":147},"PHP vulnerability PHP2Shell",[257,472,473],{},"None of these risks scales down with agency size. Two of them — technical debt\nand handover readiness — are structurally more likely in large teams, because\nnobody holds the whole system in their head.",[475,476,480],"faq-section",{":items":477,"background":478,"label":479},"[{\"question\":\"Does AI make small agencies more capable than large ones?\",\"answer\":\"Not automatically. According to DORA's research, AI acts as an amplifier of existing conditions — it magnifies the strengths of high-performing teams and the dysfunctions of weak ones. Small teams benefit where short decision paths and senior staffing come together, because the bottleneck has moved from writing to judging. A small team without a verification system is not better for it, just faster at accumulating technical debt.\"},{\"question\":\"How much faster is development with AI, really?\",\"answer\":\"What holds up is mainly the range. A randomised trial by METR found that experienced developers using early-2025 tools were 19 percent slower; the same developers with late-2025 tools came out an estimated 18 percent faster, with a wide confidence interval. Across the industry, over 80 percent of professionals report higher productivity while delivery instability rises at the same time. The gain therefore depends on practice, task type and verification system, not on the tool alone.\"},{\"question\":\"Why does an agency's headcount matter less than it used to?\",\"answer\":\"Because writing capacity is no longer the bottleneck. Coordination overhead grows quadratically with team size, and in larger organisations several handovers sit between sales and implementation. That overhead used to be offset by real capacity; since the leverage of raw implementation hours has fallen, the overhead remains without a counterweight. For parallel multi-country rollouts or 24\u002F7 operations, size is still the right answer.\"},{\"question\":\"How can I tell whether an agency uses AI responsibly?\",\"answer\":{\"By its verification system, not by its tools\":{\" Useful questions are\":\"who reads generated code back, and against which standard? Does the CI block when it is red? Are there tests that protect behavior rather than implementation? Is documentation produced that another team could take over with? And can the agency explain why a requirement was solved through standard extension points instead of custom code?\"}}},{\"question\":\"Does AI make software development cheaper?\",\"answer\":\"Rarely on the invoice, often in effect. What gets cheaper is the first working version. What remains is verification, operations, updates and the maintenance of every feature built. The real gain usually shows up not as a lower quote but as more iterations within the same budget — and as the ability to discard things before they get expensive.\"}]","default","FAQ",[257,481,482],{},"Common questions about AI in software development and choosing a provider.",[267,484,486],{"id":485},"how-to-evaluate-an-agency-today","How to evaluate an agency today",[257,488,489],{},"Headcount appears in every pitch deck because it is easy to state. These eight\nquestions are harder to answer and say considerably more:",[491,492,493,500,506,512,518,524,530,536],"ol",{},[494,495,496,499],"li",{},[275,497,498],{},"Does the person I am talking to build?"," If the answer involves a role\nthat does not work in the code, there is a handover — and with it the first\nplace where context gets lost.",[494,501,502,505],{},[275,503,504],{},"Which decision would you make against the standard in our project, and\nwhy?"," An agency that cannot offer a concrete trade-off here has not yet\nworked through it.",[494,507,508,511],{},[275,509,510],{},"Who reads generated code back?"," And in what ratio to the volume\nproduced? Review is the limiting factor, not generation.",[494,513,514,517],{},[275,515,516],{},"What do your projects test — behavior or implementation?"," Tests against\nimplementation break with every refactoring and therefore get switched off.",[494,519,520,523],{},[275,521,522],{},"Does your CI block?"," A pipeline status that is allowed to stay red is\ndecoration.",[494,525,526,529],{},[275,527,528],{},"How long does an architecture decision take with you?"," \"Hours\" is an\nanswer; \"we'll clarify that internally\" is a description of an org chart.",[494,531,532,535],{},[275,533,534],{},"What would you cut from our requirements list?"," Anyone who cuts nothing\nis selling implementation rather than outcomes.",[494,537,538,541],{},[275,539,540],{},"Could another team take over this system in six months?"," And how would\none recognise that before placing the order — documentation, tests,\narchitecture description?",[257,543,544],{},"None of these questions has anything to do with size. All eight can be\nanswered just as well by a two-person team as by an agency with 120 people —\nand in our experience often better, because the answer does not have to be\nfetched internally first.",[267,546,548],{"id":547},"what-this-means-for-budget-and-timeline","What this means for budget and timeline",[257,550,551],{},"Three things that have genuinely changed in projects:",[257,553,554,557,558,561],{},[275,555,556],{},"Prototypes have become almost free."," A contested question — configurator\nor not, this checkout flow or that one — can now be answered in days as a\nworking piece of software instead of in weeks as a concept paper. That moves\ndecisions from opinion to evidence. The numbers from a\n",[280,559,560],{"href":144},"measurable strategy"," are the\nfoundation for exactly this: without a defined metric, the prototype is just a\nnew opinion.",[257,563,564,567,568,426],{},[275,565,566],{},"The stack decision weighs more than it used to."," When implementation is\nfast, the choice of platform becomes the dominant cost factor over the\nlifetime. Licence model, fee structure, extensibility and upgrade path decide\nthings for years — and are hard to correct. We worked that trade-off through\nin detail for two platforms:\n",[280,569,570],{"href":168},"Shopware or Shopify",[257,572,573,576],{},[275,574,575],{},"The outcome share of the budget rises."," If less budget goes into\nimplementation, more can go into analysis, measurement and iteration — or it\ngets spent on additional features. That is a decision, not a consequence. It\nis made during prioritisation, and that is exactly where the difference lies\nbetween a project that delivers and one that contains a lot.",[257,578,579],{},"What has not changed: a project does not become half as expensive because the\nmodels got better. The effort moves from production into judgement — and\njudgement is senior work.",[267,581,583],{"id":582},"how-we-hold-this-line","How we hold this line",[257,585,586,587,590],{},"We are two founders, and for us that is not a growth stage but a decision:\n",[280,588,589],{"href":81},"output per head instead of heads per project",". Whoever advises also\nbuilds. There is no handover between the conversation and the code, and\narchitecture decisions are made in hours rather than through approval chains.",[257,592,593],{},"Five years ago that structure was a capacity risk. With AI tools compressing\nthe implementation share, it turns into an advantage — but only together with\nthe part that remains work: tests that protect behavior, a CI that blocks,\nmonitoring from day one, documentation another team could take over with. We\nuse AI daily, and we use it where it holds: accelerating implementation, not\nreplacing decisions. Which decision gets made is owned by a person with a\nname.",[257,595,596,597,600],{},"How that fits together methodically is described in\n",[280,598,599],{"href":18},"F.O.C.U.S."," — Facts before Objectives, Constraints\nbefore Execution. With AI, the order has become more important, not more\noptional: anyone who can build faster can also build the wrong thing faster.",[267,602,604],{"id":603},"the-next-step","The next step",[257,606,607],{},"Three checks you do not need a pitch for:",[491,609,610,616,622],{},[494,611,612,615],{},[275,613,614],{},"Ask your current provider who reads generated code back."," The answer\ntells you more about your risk than any reference list.",[494,617,618,621],{},[275,619,620],{},"Take the last three rejected features and recalculate them."," If\nimplementation now costs a fraction, that changes prioritisation — or it\nshows they were rejected for good reasons.",[494,623,624,627],{},[275,625,626],{},"Check whether your system is handover-ready."," Not because you want to\nswitch, but because the answer is the best indicator of its condition.",[257,629,630,631,634,635,638,639,642],{},"If you would rather not make the stack and implementation decisions alone:\nthat is exactly what we do in ",[280,632,633],{"href":90},"consulting",", and in\n",[280,636,637],{"href":93},"development"," we implement them.\n",[280,640,641],{"href":30},"Get in touch"," — a conversation with the people who also build.",[267,644,646],{"id":645},"sources-and-further-reading","Sources and further reading",[648,649,650,656,663,670,676,682,688],"ul",{},[494,651,652],{},[280,653,655],{"href":326,"rel":654},[284],"DORA (Google Cloud): Balancing AI tensions — Moving from AI adoption to effective SDLC use",[494,657,658],{},[280,659,662],{"href":660,"rel":661},"https:\u002F\u002Fservices.google.com\u002Ffh\u002Ffiles\u002Fmisc\u002F2025_state_of_ai_assisted_software_development.pdf",[284],"DORA: State of AI-assisted Software Development 2025 (PDF)",[494,664,665],{},[280,666,669],{"href":667,"rel":668},"https:\u002F\u002Fcloud.google.com\u002Fresources\u002Fcontent\u002Fdora-roi-of-ai-assisted-software-development",[284],"DORA: ROI of AI-assisted Software Development",[494,671,672],{},[280,673,675],{"href":282,"rel":674},[284],"METR: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity",[494,677,678],{},[280,679,681],{"href":304,"rel":680},[284],"METR: We are Changing our Developer Productivity Experiment Design (February 2026)",[494,683,684],{},[280,685,687],{"href":338,"rel":686},[284],"Stack Overflow Developer Survey 2026: Trust in AI tools",[494,689,690],{},[280,691,693],{"href":395,"rel":692},[284],"Shopware Developer Docs: Plugin Base Guide",[267,695,697],{"id":696},"read-on","Read on",[648,699,700,705,710,715,720],{},[494,701,702],{},[280,703,704],{"href":168},"Shopware or Shopify: platform comparison 2026",[494,706,707],{},[280,708,709],{"href":171},"Maintenance, retainers and continuous development",[494,711,712],{},[280,713,714],{"href":144},"Measurable strategy in e-commerce",[494,716,717],{},[280,718,719],{"href":18},"F.O.C.U.S. — the methodology behind our projects",[494,721,722],{},[280,723,724],{"href":93},"Development: build, don't discuss",{"title":726,"searchDepth":727,"depth":727,"links":728},"",2,[729,734,735,736,742,743,744,745,746,747,748],{"id":269,"depth":727,"text":270,"children":730},[731,733],{"id":293,"depth":732,"text":294},3,{"id":315,"depth":732,"text":316},{"id":344,"depth":727,"text":345},{"id":357,"depth":727,"text":358},{"id":378,"depth":727,"text":379,"children":737},[738,739,740,741],{"id":385,"depth":732,"text":386},{"id":401,"depth":732,"text":402},{"id":411,"depth":732,"text":412},{"id":429,"depth":732,"text":430},{"id":439,"depth":727,"text":440},{"id":485,"depth":727,"text":486},{"id":547,"depth":727,"text":548},{"id":582,"depth":727,"text":583},{"id":603,"depth":727,"text":604},{"id":645,"depth":727,"text":646},{"id":696,"depth":727,"text":697},[750,751,752,753,754,755,756],{"name":655,"url":326},{"name":662,"url":660},{"name":669,"url":667},{"name":675,"url":282},{"name":681,"url":304},{"name":687,"url":338},{"name":693,"url":395},"AI moved the bottleneck in software development — from writing code to judging it. Why an agency's headcount no longer says anything about project speed, and which decisions count instead.",false,"md",{"src":761,"alt":762},"\u002Fimages\u002Finsights\u002Fki-softwareentwicklung-agenturgroesse.png","Benedikt Rillox shows how the bottleneck in software development moved from writing code to judging it — a large team with many coordination paths next to a small senior team with a short decision chain",{},true,null,"2026-10-28",14,{"title":769,"description":770,"articleModifiedTime":766},"AI in software development: agency size stopped mattering","Why project speed depends on tech stack decisions and review systems instead of headcount — with data from DORA, METR and Stack Overflow 2026.",{"loc":138,"lastmod":766},"en\u002Finsights\u002Fai-software-development-agency-size",[774,775,776,777],"AI","Technology","Strategy","Agency Selection","insights-ai-agency-size","aYqah9pXRxWrIn49u1uiQC7dkR8y9HR_yixFa0whSHo",{"id":781,"title":782,"alternateName":765,"bio":783,"body":784,"description":954,"events":955,"extension":759,"heroIllustration":957,"image":959,"meta":962,"name":961,"navigation":764,"ogImage":765,"path":243,"projects":765,"publishedAt":765,"robots":765,"role":963,"sameAs":964,"schemaOrg":765,"seo":968,"sitemap":970,"slug":252,"socials":971,"squeeze":758,"stem":972,"talks":765,"translationKey":973,"updatedAt":765,"__hash__":974},"team_en\u002Fen\u002Fteam\u002Fbenedikt-rillox.md","Benedikt Rillox — Founder & Strategist","Benedikt is managing partner of nuonic Digital, e-commerce consultant, and developer. Since 2009 he has been designing, building, and guiding digital platforms — from code through product ownership to strategic leadership.",{"type":254,"value":785,"toc":945},[786,790,796,799,802,805,809,812,815,818,821,824,828,831,834,837,860,863,867,870,873,876,893,897,900,903,909,912,916,919,922,925,929,932,935],[267,787,789],{"id":788},"about-benedikt-rillox","About Benedikt Rillox",[791,792,793],"blockquote",{},[257,794,795],{},"To me, technology is not a cost item — it's a competitive advantage.",[257,797,798],{},"I'm Benedikt Rillox, managing partner of nuonic Digital, e-commerce consultant, and developer.",[257,800,801],{},"Since 2009 I have been designing, building, and guiding digital platforms. My path led from hands-on software development through product ownership and technical team leadership into entrepreneurial responsibility.",[257,803,804],{},"That's why I know e-commerce from very different perspectives: from the code, from product development, from budget responsibility, and from strategic leadership. Today I combine these experiences to build commerce systems that don't just work technically, but measurably move businesses forward.",[267,806,808],{"id":807},"my-path-in-e-commerce","My path in e-commerce",[257,810,811],{},"I started out as a developer for web and e-commerce projects using technologies like Shopware, TYPO3, PHP, and Symfony. During that time I delivered solutions for smaller companies as well as for well-known brands and corporations — including Unilever, Paul Hewitt, and Framily.",[257,813,814],{},"At A-Z Gartenhaus my role evolved from Shopware developer to product owner. There I helped build and lead a ten-person development team and was responsible for the strategic evolution of the commerce platform and its connected third-party systems.",[257,816,817],{},"For the Legero United group I took on budget responsibility as product owner and coordinated the technical development of the Superfit, legero, and Think! shops.",[257,819,820],{},"As tech lead for ByteClub I planned and built the commerce platforms of the group with brands like COMSPOT, Shifter, GfBD, and IM.F.B. Further projects included ongoing support and the Shopware 6 rebuild for Xucker, as well as the design and development of a Shopware payment plugin for topi that digitizes B2B leasing and buy-on-invoice.",[257,822,823],{},"These stations taught me that successful commerce projects never consist of technology alone. They combine architecture, processes, people, and clear business goals.",[267,825,827],{"id":826},"from-development-to-strategy","From development to strategy",[257,829,830],{},"I am a developer, a strategist, and a problem solver.",[257,832,833],{},"I can dive deep into technical details without losing sight of the actual goal: What impact should a solution have for the company, its people, and its customers?",[257,835,836],{},"Today my focus is on designing, building, and continuously evolving demanding Shopware systems. Among other things, that includes:",[648,838,839,842,845,848,851,854,857],{},[494,840,841],{},"technical and strategic e-commerce consulting,",[494,843,844],{},"software and Shopware development,",[494,846,847],{},"architecture and system integration,",[494,849,850],{},"product ownership and project management,",[494,852,853],{},"performance and stable operations,",[494,855,856],{},"automation and process optimization,",[494,858,859],{},"technical team leadership and collaboration with agencies.",[257,861,862],{},"Across more than 25 client projects I have seen what makes digital initiatives fail — and what makes them succeed. Good technology needs a clear direction, realistic requirements, and the courage to make decisions.",[267,864,866],{"id":865},"i-dont-build-shops-just-so-theyre-online","I don't build shops just so they're online",[257,868,869],{},"A shop can work technically and still be too slow, too complicated, or economically ineffective. That's why at nuonic we don't just look at individual features. We look at the whole system: architecture, processes, integrations, performance, customer journey, and the ability to grow in the future.",[257,871,872],{},"I don't believe in best practices when they merely mean copying what everyone else is already doing. I believe in solutions that fit a company's business model, its team, and its actual goals.",[257,874,875],{},"That also means asking uncomfortable questions:",[648,877,878,881,884,887,890],{},[494,879,880],{},"Does this feature really need to exist?",[494,882,883],{},"Does it solve a relevant problem?",[494,885,886],{},"Will the system remain manageable in the long run?",[494,888,889],{},"Does the investment support a concrete business goal?",[494,891,892],{},"Can the platform grow without collapsing under its own complexity?",[267,894,896],{"id":895},"responsibility-begins-after-go-live","Responsibility begins after go-live",[257,898,899],{},"To me, go-live is not a success. It's the beginning of responsibility.",[257,901,902],{},"Only in day-to-day operations does a platform prove whether it truly holds up: whether it stays stable, whether the team can work with it efficiently, and whether it can evolve without unnecessary friction.",[257,904,905,906],{},"That's why my work doesn't end with the handover. That's where the real value creation begins: ",[275,907,908],{},"Optimize. Automate. Scale.",[257,910,911],{},"I don't think in short-term projects, but in commerce systems that carry their weight for the long run. Technology shouldn't just cover today's needs — it should open up new possibilities.",[267,913,915],{"id":914},"clarity-is-part-of-working-together","Clarity is part of working together",[257,917,918],{},"Good results don't come from saying yes to every idea. That's why I also say no — to bad requirements, to short-term thinking, and to projects without a discernible direction.",[257,920,921],{},"I prefer working with decision-makers who take ownership. With teams that are ready to genuinely change things instead of merely managing existing problems. And with companies that see technology as an investment — not as a necessary cost item.",[257,923,924],{},"That's not always comfortable. But it creates clarity, protects budgets, and leads to better results.",[267,926,928],{"id":927},"what-i-stand-for","What I stand for",[257,930,931],{},"I don't want to simply be busy. I want to create impact.",[257,933,934],{},"With nuonic we build commerce systems for companies that want to grow — faster, more stable, and more profitable.",[257,936,937,938,941,942,944],{},"Others talk about digitalization. We deliver it.",[939,940],"br",{},"\nOthers plan. We build.",[939,943],{},"\nOthers wait. We go.",{"title":726,"searchDepth":727,"depth":727,"links":946},[947,948,949,950,951,952,953],{"id":788,"depth":727,"text":789},{"id":807,"depth":727,"text":808},{"id":826,"depth":727,"text":827},{"id":865,"depth":727,"text":866},{"id":895,"depth":727,"text":896},{"id":914,"depth":727,"text":915},{"id":927,"depth":727,"text":928},"Projects, references, talks and events by Benedikt Rillox — founder and strategist at nuonic Digital, e-commerce consultant and developer since 2009.",[956],"omr-festival-2026",{"src":958,"alt":726},"\u002Fimages\u002Fteam\u002Fdefender.png",{"src":960,"alt":961},"\u002Fimages\u002Fteam\u002Fbenedikt-rillox.jpg","Benedikt Rillox",{},"Founder & Strategist",[965,966,967],"https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fbenedikt-rillox-4122b183\u002F","https:\u002F\u002Fgithub.com\u002Fbenediktrillox","https:\u002F\u002Fwww.xing.com\u002Fprofile\u002FBenedikt_Rillox",{"title":782,"description":954,"articleModifiedTime":969},"2026-08-14",{"loc":243,"lastmod":969},{"linkedin":965,"github":966},"en\u002Fteam\u002Fbenedikt-rillox","team-benedikt-rillox","Io4iBFT4BZaQMrk8v8Woi1nCjxdLOcoLJReSh5tJVUg",1793208823136]