[{"data":1,"prerenderedAt":13},["ShallowReactive",2],{"$f1l7cpetkavo7o":3},{"title":4,"description":5,"canonicalPath":6,"html":7,"kind":8,"author":9,"image":10,"publishedTime":11,"modifiedTime":12},"Preparing Your Business for AI Agents | Irvine-David","Learn what data, permissions, integrations, business rules and operational controls must be ready before connecting AI agents to your organisation.","\u002Fblog\u002Fthe-agent-is-the-easy-bit-what-needs-to-be-in-place-before-you-connect-ai-to-your-business","\u003Csection class=\"page-hero blog-page-hero\" style=\"--page-hero-image:url('\u002Fimages\u002Fblog\u002Fthe-agent-is-the-easy-bit-what-needs-to-be-in-place-before-you-connect-ai-to-your-business.jpg')\">\n    \u003Cdiv class=\"wrap\">\n      \u003Cnav class=\"eyebrow page-trail\" aria-label=\"Breadcrumb\">\u003Cspan class=\"dot\">\u003C\u002Fspan>\u003Ca href=\"\u002F\">Home\u003C\u002Fa>\u003Cspan aria-hidden=\"true\">\u002F\u003C\u002Fspan>\u003Ca href=\"\u002Fblog\">Insights\u003C\u002Fa>\u003C\u002Fnav>\n      \u003Ch1 style=\"margin-top:18px\">The Agent Is the Easy Bit: What Needs to Be in Place Before You Connect AI to Your Business?\u003C\u002Fh1>\n      \u003Cp class=\"blog-meta\">16 September 2026\u003C\u002Fp>\n   \u003C\u002Fdiv>\n  \u003C\u002Fsection>\n  \u003Carticle class=\"content-section blog-article\">\n    \u003Cdiv class=\"wrap blog-body editorial-body\">\u003Cdiv class=\"editorial-intro\">\u003Cp>AI agents are becoming surprisingly easy to build.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">With tools such as Microsoft Copilot Studio, Azure AI and modern development frameworks, it is now possible to create an agent that can understand a request, reason about what to do next, search for information, call APIs and perform actions.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The technology is moving incredibly quickly.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">But in a real business, \u003Cstrong>the agent is often the easy bit. The surrounding business architecture is where the real work is.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Before an AI agent can safely do anything genuinely useful, it needs reliable access to the right data, appropriate permissions, secure authentication, well-defined integrations and clear rules around what it is — and is not — allowed to do.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">It also needs to operate within the reality of an existing organisation.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That might mean SQL Server, Dataverse, SharePoint, Microsoft 365, Power Platform, bespoke applications, third-party platforms and legacy systems that have evolved over many years.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">This is why successful AI agent projects are rarely just AI projects.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">They are usually \u003Cstrong>data, integration, security and software architecture projects with AI sitting on top\u003C\u002Fstrong>.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Getting those foundations right is often the difference between an impressive demonstration and an AI agent that can genuinely be trusted in production.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Moving Beyond the Chatbot\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">For many organisations, the first experience of generative AI was conversational.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Ask a question.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Get an answer.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Generate some content.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Summarise a document.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That remains useful, but an AI agent can potentially go much further.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent might be able to:\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cp>find information across company systems;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>identify a customer or client;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>retrieve their recent activity;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>analyse documents;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>update a CRM record;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>create a task;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>prepare a report;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>call an internal API;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>trigger a workflow;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>recommend a next action;\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>or perform that action itself.\u003C\u002Fp>\u003C\u002Fli>\u003C\u002Ful>\u003Cp class=\"isSelectedEnd\">That is a significant change.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The moment AI moves from \u003Cstrong>answering questions\u003C\u002Fstrong> to \u003Cstrong>performing actions\u003C\u002Fstrong>, architecture becomes much more important.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An incorrect answer from a chatbot may be inconvenient.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent incorrectly updating a customer record, sending a communication, approving something or triggering a business process can have considerably greater consequences.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The question therefore stops being simply:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Can the AI do this?”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">It becomes:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Under what circumstances should we allow the AI to do this?”\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Start With the Data\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">Most useful business agents depend on company data.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That sounds obvious, but it immediately raises some difficult questions.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Where is the information?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Is it in Dataverse?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">SQL Server?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">SharePoint?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A line-of-business application?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Several systems?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Is the information current?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Is there a reliable way of identifying the same customer, employee or transaction across those systems?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Which system is the authoritative source?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Can the data be accessed through an API?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">And perhaps most importantly: can you trust it?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Giving an AI agent access to disorganised or inconsistent information does not magically make that information better.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">In some cases it simply allows the organisation to access bad data more efficiently.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Before introducing an agent, it can therefore be valuable to improve the underlying data architecture first.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That might involve creating better APIs, consolidating information, introducing clearer data models, improving master data or establishing which system owns which information.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI tends to expose these architectural weaknesses very quickly.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Identity and Permissions Matter\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">An agent should not automatically have access to everything simply because it is technically capable of accessing it.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The same security principles that apply to normal business applications should apply to AI.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If an employee asks an agent about a customer, should the agent be allowed to retrieve every document associated with that customer?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Should everybody be able to see salary information?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Commercial information?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Board documents?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Personal data?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Financial records?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If a user cannot normally access something, using an AI agent should not become a way of bypassing that restriction.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">This means understanding identity and permissions properly.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">In a Microsoft environment, that may involve Microsoft Entra ID, Dataverse security roles, SharePoint permissions, application identities, API permissions and the security models of individual systems.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent also needs its own identity when communicating with other services.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Those permissions should normally be limited to what the agent genuinely needs.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Giving an agent unrestricted administrator access because it makes the initial integration easier is not a sensible production architecture.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Give the Agent Tools, Not Unlimited Access\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">One of the most important architectural decisions is how an agent interacts with business systems.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">There is a significant difference between giving an AI model unrestricted access to a database and giving it a carefully designed tool that performs a specific operation.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Consider an agent working with customer information.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Rather than allowing it to construct arbitrary database commands, the system might expose controlled capabilities such as:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">GetCustomer\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">GetRecentInteractions\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">CreateFollowUpTask\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">UpdateCustomerAddress\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">RequestDocument\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Each capability can then have its own validation, security rules and audit trail.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The agent can decide \u003Cstrong>which tool it needs to use\u003C\u002Fstrong>, while traditional software remains responsible for making sure that tool behaves safely.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">This combination can be extremely powerful.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI handles the interpretation and reasoning.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Normal software handles deterministic business rules.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That separation is important.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Not every decision needs AI.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">In fact, wherever a rule can be expressed reliably in normal software, it is often better to keep it there.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Your Existing APIs Become Increasingly Valuable\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">Businesses that already have well-designed APIs are often in a much stronger position to introduce AI agents.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An API effectively provides a controlled contract between the agent and the underlying system.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The agent does not necessarily need to understand how an application works internally.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">It simply needs an approved way of asking it to do something.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That could be:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Find this client.”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Retrieve their portfolio.”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Create an activity.”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Calculate this figure.”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Generate a quotation.”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Schedule a review.”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">This is one reason we believe traditional software engineering becomes more important rather than less important as AI develops.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI may change how users interact with systems, but those systems still need secure and reliable ways of exposing their capabilities.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Modern approaches such as the Model Context Protocol (MCP) are making it easier for AI systems to discover and work with tools.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">But the underlying principle remains familiar.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A well-designed interface between systems is valuable.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\u003Cfigure class=\"editorial-figure\">\n          \u003Cimg src=\"\u002Fimages\u002Fblog\u002Fthe-agent-is-the-easy-bit-what-needs-to-be-in-place-before-you-connect-ai-to-your-business.jpg\" alt=\"\" loading=\"lazy\" \u002F>\n          \u003Cfigcaption>Ideas become valuable when they connect to real business context.\u003C\u002Ffigcaption>\n        \u003C\u002Ffigure>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Business Rules Still Belong Somewhere\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">Businesses often contain years of accumulated knowledge in their software.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A customer may only be eligible for a particular product under certain circumstances.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A transaction above a certain value may require approval.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A workflow may behave differently depending on the client's status.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A calculation may depend on several pieces of information.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">These rules should not simply disappear into an AI prompt.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Where rules are important, repeatable and deterministic, they should generally remain within normal application logic.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The agent can decide that a calculation needs to be performed.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A trusted service should perform the calculation.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The agent can decide that a workflow should be initiated.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The workflow should still enforce the organisation's normal rules.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI is excellent at dealing with ambiguity.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Software is excellent at enforcing certainty.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Good agent architecture uses both.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Decide Where Humans Remain in Control\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">One of the biggest design questions is how much autonomy an agent should have.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">There is no single answer.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The appropriate level depends on the action being performed.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent that summarises internal information might need very little human intervention.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent preparing a draft email may simply require the employee to review it before sending.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent proposing changes to a client record might require confirmation.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent initiating a significant financial transaction should probably have considerably stronger controls.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A useful way to think about this is as a progression:\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Inform → Recommend → Prepare → Approve → Act\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">The agent might initially only provide information.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Later it could recommend an action.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Then it could prepare that action for a human to approve.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Only when the organisation has sufficient confidence in the process might the agent be allowed to act autonomously.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI adoption does not need to jump directly from chatbot to autonomous business process.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Gradual autonomy can be a much more sensible route.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>What Happens When the Agent Gets It Wrong?\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">This is one of the most important questions in any production AI system.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Not:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Will it ever make a mistake?”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">But:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“What happens when it does?”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A production architecture should assume that errors will occasionally occur.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That means thinking about auditability.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">What did the user ask?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">What information did the agent retrieve?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Which tools did it call?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">What actions did it perform?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">What changed as a result?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Can those actions be reversed?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Can somebody investigate what happened?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Traditional business systems have spent decades developing concepts such as logging, transaction histories, permissions, approval workflows and audit trails.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI systems need the same level of operational thinking.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">An agent should not become an invisible layer performing actions across an organisation without leaving a clear record of what happened.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Monitoring Matters Too\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">An AI agent is not something that should simply be deployed and forgotten.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Its behaviour needs to be monitored.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Are people using it?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">What are they asking?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Which questions does it struggle with?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Which tools is it calling?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Where does it fail?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Are there recurring requests that suggest a missing integration?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Is it producing useful outcomes?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Is it creating unnecessary work?\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The answers can help improve both the agent and the wider business systems surrounding it.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Interestingly, AI agents may also highlight weaknesses in existing processes.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If an agent repeatedly struggles to find the correct information, perhaps employees struggle too.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If three systems need to be searched before a simple customer question can be answered, perhaps the real problem is the architecture rather than the agent.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI can become a useful lens through which to examine the organisation itself.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\u003Caside class=\"editorial-cta\" aria-label=\"Explore this topic\">\n      \u003Cdiv>\n        \u003Cspan>Put the thinking into practice\u003C\u002Fspan>\n        \u003Cstrong>What could this mean for your business?\u003C\u002Fstrong>\n      \u003C\u002Fdiv>\n      \u003Ca href=\"\u002Fexpertise\u002Fai-agents\">Technology expertise →\u003C\u002Fa>\n    \u003C\u002Faside>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Not Everything Needs an Agent\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">There is understandably a great deal of enthusiasm around AI agents.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">But not every business process needs one.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If a process is completely predictable, a conventional workflow may be faster, cheaper and safer.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If something always happens when a record changes, Power Automate may be the correct answer.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If a complex calculation needs to be performed reliably, a .NET service may be appropriate.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If users simply need better reporting, Power BI may solve the problem.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Agents become particularly interesting when the task involves ambiguity.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Understanding what someone is asking.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Gathering information from several places.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Choosing between different tools.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Interpreting documents.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Applying context.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Recommending what should happen next.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The objective should not be to insert AI into every process.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">It should be to use AI where it adds something that traditional software struggles to provide.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>The Best Agent Architecture Is Usually Hybrid\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">In practice, a useful business agent will often sit across several technologies.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Imagine an employee asking:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">“Can you prepare everything we need for Sarah's annual client review?”\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Behind that relatively simple question, quite a lot might happen.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The agent may identify the correct client in Dataverse.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A .NET API might retrieve information from a specialist business application.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Documents might come from SharePoint.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Financial information might sit in SQL Server.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A Power Automate flow might create internal tasks.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The agent may summarise recent correspondence and identify missing information.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A reporting platform might provide additional analysis.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Finally, the employee receives a concise summary and a list of proposed actions.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">To the user, it feels like one intelligent assistant.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Behind the scenes, it is an architecture.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">And that architecture is what makes the agent useful.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>Before You Build an Agent, Ask These Questions\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">Before connecting an AI agent to your organisation, it is worth asking:\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cp>What business problem are we actually trying to solve?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>Where does the agent need to obtain its information?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>Can we trust that information?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>Which systems does it need to interact with?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>Do those systems have suitable APIs or integration points?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>How will users and the agent authenticate?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>What should each user be allowed to see?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>Which actions can the agent perform?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>Which actions require human approval?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>Which business rules must remain deterministic?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>How will actions be logged and audited?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>What happens when something goes wrong?\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>How will we monitor whether the agent is actually creating value?\u003C\u002Fp>\u003C\u002Fli>\u003C\u002Ful>\u003Cp class=\"isSelectedEnd\">Those questions are considerably less exciting than asking which AI model to use.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">But they are often much more important.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>AI Agents Make Good Architecture More Valuable\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">There is an interesting consequence to the rapid development of AI.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">As agents become easier to build, the quality of the systems surrounding them becomes increasingly important.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A business with structured data, modern APIs, well-defined permissions, sensible integration patterns and clear business processes can experiment with AI relatively quickly.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">A business with fragmented systems, inconsistent data and unclear ownership may discover that the agent itself is the least of its problems.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That does not mean everything needs to be modernised before an organisation can start experimenting with AI.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">It does mean the experiment should help identify the architecture that will eventually be required.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Start small.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Give the agent access to a clearly defined set of information.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Introduce a small number of useful tools.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Keep humans involved in important decisions.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Monitor what happens.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Then expand its capabilities as confidence grows.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>The Agent Really Is the Easy Bit\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">Creating an impressive AI demonstration has never been easier.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Creating an AI capability that employees can rely on every day is a different challenge.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That requires much of the same thinking that has always gone into good business technology:\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">data architecture;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">security;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">identity;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">permissions;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">integration;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">APIs;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">business rules;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">monitoring;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">governance;\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">and good software engineering.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">AI does not remove the need for those disciplines.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">It makes them more important.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">The organisations that gain the most from AI agents are unlikely to be those that simply deploy the largest number of them.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">They will be the organisations that connect AI intelligently to the data, software and processes that already make their business work.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">\u003Cstrong>The agent may be the exciting part.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">\u003Cstrong>But the architecture around it is what turns it into a business system.\u003C\u002Fstrong>\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n        \u003Csection class=\"editorial-section editorial-section--panel\">\n          \u003Cdiv class=\"editorial-section-heading\">\u003Ch2>How Irvine-David Can Help\u003C\u002Fh2>\u003C\u002Fdiv>\n          \u003Cdiv class=\"editorial-section-copy\">\u003Cp class=\"isSelectedEnd\">At Irvine-David, we help businesses look beyond the AI demonstration and consider the wider technology required to make AI genuinely useful.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">That can include reviewing existing applications and data, designing APIs and integrations, working with \u003Ca href=\"\u002Fexpertise\u002Fdata-platforms\">Microsoft Power Platform and Dataverse\u003C\u002Fa>, integrating SQL Server and SharePoint, developing bespoke .NET services, implementing AI agents and establishing the security and governance required around them.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">Because our work spans \u003Ca href=\"\u002Fexpertise\u002Fmicrosoft-development\">software development\u003C\u002Fa>, Microsoft technologies, data platforms, AI and \u003Ca href=\"\u002Fhow-we-help\u002Fowner-ai-technology-partner\">technology strategy\u003C\u002Fa>, we can look at the agent as part of the wider business architecture rather than as a standalone tool.\u003C\u002Fp>\u003Cp class=\"isSelectedEnd\">If you are considering introducing AI agents into your business, a useful first step is not necessarily deciding which agent platform to use.\u003C\u002Fp>\u003Cp>It is understanding what you want the agent to do — and whether the systems around it are ready to let it do it safely.\u003C\u002Fp>\u003C\u002Fdiv>\n        \u003C\u002Fsection>\n    \u003Caside class=\"editorial-related\" aria-label=\"Related reading\">\n      \u003Cdiv class=\"editorial-related-heading\">\n        \u003Cspan class=\"dot\">\u003C\u002Fspan>\n        \u003Cstrong>Continue exploring\u003C\u002Fstrong>\n      \u003C\u002Fdiv>\n      \u003Cdiv class=\"editorial-related-grid\">\n        \u003Ca href=\"\u002Fexpertise\u002Fai-agents\">\u003Cspan>Technology expertise\u003C\u002Fspan>\u003Cstrong>Connect AI agents to trusted data, rules and business systems\u003C\u002Fstrong>\u003Cb aria-hidden=\"true\">↗\u003C\u002Fb>\u003C\u002Fa>\u003Ca href=\"\u002Fhow-we-help\u002Fai-opportunity-review\">\u003Cspan>How we help\u003C\u002Fspan>\u003Cstrong>Identify the AI opportunities worth pursuing first\u003C\u002Fstrong>\u003Cb aria-hidden=\"true\">↗\u003C\u002Fb>\u003C\u002Fa>\n      \u003C\u002Fdiv>\n    \u003C\u002Faside>\u003Cdiv class=\"blog-article-footer\">\n        \u003Ca href=\"\u002Fblog\">← Back to all insights\u003C\u002Fa>\n     \u003C\u002Fdiv>\n   \u003C\u002Fdiv>\n  \u003C\u002Farticle>\n  \u003Csection class=\"content-pagination-section\">\n  \u003Cdiv class=\"wrap\">\n    \u003Ca class=\"content-index-link\" href=\"\u002Fblog\">← All insights\u003C\u002Fa>\n    \u003Cnav class=\"content-pagination\" aria-label=\"Article navigation\">\n      \u003Ca href=\"\u002Fblog\u002Flegacy-software-modernisation\">\u003Cspan>Previous\u003C\u002Fspan>\u003Cstrong>Your Business Runs on Software Nobody Fully Understands. What Happens Next?\u003C\u002Fstrong>\u003C\u002Fa>\n      \u003Ca class=\"next\" href=\"\u002Fblog\u002Fwhat-does-a-fractional-cto-do-and-does-your-business-need-one\">\u003Cspan>Next\u003C\u002Fspan>\u003Cstrong>What Does a Fractional CTO Do — and Does Your Business Need One?\u003C\u002Fstrong>\u003C\u002Fa>\n    \u003C\u002Fnav>\n  \u003C\u002Fdiv>\n\u003C\u002Fsection>\n\n\u003Csection class=\"next-bar\">\n    \u003Cdiv class=\"wrap next-flex\">\n      \u003Cdiv>\n        \u003Ch3>Have a technology question of your own?\u003C\u002Fh3>\n        \u003Cp class=\"section-lead\">Bring us the business problem and we can explore the right next step.\u003C\u002Fp>\n     \u003C\u002Fdiv>\n      \u003Ca class=\"btn btn-dark\" href=\"\u002Fcontact\">Start a conversation →\u003C\u002Fa>\n   \u003C\u002Fdiv>\n  \u003C\u002Fsection>","article","Rebecca Pukiello-Slack","https:\u002F\u002Fwww.irvine-david.co.uk\u002Fimages\u002Fblog\u002Fthe-agent-is-the-easy-bit-what-needs-to-be-in-place-before-you-connect-ai-to-your-business.jpg","2026-09-16T21:57:18+0100","2026-09-16T21:57:20+0100",1795271792479]