The latest articles on DEV Community by Anton Loginov (@antonloginovv). https://dev.to/antonloginovv https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3933965%2F431b5ac7-4f16-4f83-bbe1-0d6518d9149a.png https://dev.to/antonloginovv en Anton Loginov Fri, 15 May 2026 23:25:08 +0000 https://dev.to/antonloginovv/deploying-agents-is-harder-than-deploying-software-and-that-is-the-point-481b https://dev.to/antonloginovv/deploying-agents-is-harder-than-deploying-software-and-that-is-the-point-481b <p>A take from the delivery side, after two years shipping agentic workflows into professional-services firms.</p> <h2> Software vs agents </h2> <p>Software generally works the same way every time. For decades it has mostly been updated versions of an existing concept, so the enterprise knows how to slot it into a workflow.</p> <p>Agents are different. You are deploying the equivalent of <em>work output</em> inside the business. The customer effectively uses you as a professional-services provider for a task they expect solved nearly end to end.</p> <h2> The harder gap </h2> <p>Everyone frames this as a vendor-skill gap — you need forward-deployed engineers who understand the business. True. But the harder gap is on the customer side: most enterprise processes were never designed to be handed to <em>anyone</em>, human or agent.</p> <p>The first job on the ground is usually not deploying the agent. It is making the operating model legible enough that an agent can even be pointed at it.</p> <h2> The math nobody likes </h2> <p>Deploy is a week. The redesign is the quarter. Teams that budget only for the deploy keep shipping capable agents into processes that quietly reject them, then blame the model.</p> <p>The model is the easy part. It always was.</p> ai Anton Loginov Fri, 15 May 2026 23:24:30 +0000 https://dev.to/antonloginovv/your-ai-rollout-is-theatre-until-the-operating-model-changes-4je2 https://dev.to/antonloginovv/your-ai-rollout-is-theatre-until-the-operating-model-changes-4je2 <p>In professional services the model is the easy part. The redesign of the operating model is the work.</p> <h2> The pattern </h2> <p>Most "AI rollouts" I see fail the same way. A firm licenses a chatbot, bolts it onto a process nobody made legible first, watches nothing move for six months, then concludes AI does not apply to their kind of business.</p> <p>It is not the chatbot. It is the operating model.</p> <h2> Where the time actually goes </h2> <p>The real bottleneck was never answering questions. It is intake then triage then handoff. The moment a lead lands and a senior person spends forty minutes scoping it. The third week of an engagement when a junior re-summarises the same documents for the third meeting.</p> <p>That is not a chatbot problem. That is an agent problem. Different architecture, different ROI math.</p> <h2> What works </h2> <p>When you replace structured intellectual labour with well-designed agents that have memory, tools and human checkpoints, the numbers start to work. Deploy is a week. The redesign is the quarter. Vendors who only staff the deploy keep shipping agents into processes that quietly reject them.</p> <p>If someone told you AI does not work for professional services — you were sold the wrong product.</p> ai