Blog

<p>Pretty much exactly a year ago Des Train AI launched with it's objective of providing custom AI agents for business processes. In that time, the single most consistent request we've got is "we want an AI agent". However, it tells us almost nothing about what to build. When solving a custom business process problem, an agent is not a product you pick off a shelf. It's an architecture you design around a specific problem, a specific user, and a specific set of systems. The questions below are the ones we've learned to ask before touching a keyboard.</p><p>One thing worth clarifying upfront, because it shapes everything else: AI agents are not chatbots. A chatbot responds. An agent acts — it plans, calls tools, handles dependencies, and completes tasks that often take minutes or even hours to resolve. That's long by AI standards, but still a fraction of what a human would need to do the same work. Keep that in mind as you read through the questions. Some of them only make sense once you understand what you're actually asking the agent to do. Read more from our previous post <a data-mce-href="https://destrain.ai/blog/what-is-an-ai-agent--the-definition" href="https://destrain.ai/blog/what-is-an-ai-agent--the-definition">What Is an AI Agent? The Definition</a><br></p><h2>1. Where does it live?</h2><p>Will this agent run on someone's local laptop, or will it be deployed as a reusable service? A local agent is much more simpler and faster to prototype but it scales poorly and might be hard to maintain.&nbsp; A server-side deployment opens up multi-user access, better observability, and integration with enterprise systems, but it requires thinking seriously about reliability, cost, and security from day one. The answer here shapes almost every architectural decision that follows. We ask this first because we've seen projects get months in before discovering the deployment model doesn't match the use case. Ultimately boils down to basic software processes, organizational guidelines and architecture, not so much on the agent itself.</p><h2>2. How many users will it serve?</h2><p>An agent built for one internal analyst is a very different system from one built for ten thousand customers. With proper archtecture we can extend simplicity flexibly and cheap to few persons, even a small team, but as soon as the amount of users soar it requires different level of scalability. Scale affects not just compute and cost, but how you handle rate limits on third-party tools, how you queue and prioritize tasks, and whether you need to isolate user contexts from each other. The gap between "a few internal users" and "customer-facing at scale" is larger than most clients expect when they first describe their requirements.</p><h2>3. How does the user authenticate and how does the agent authenticate to its tools?</h2><p>This is where more agent projects run into trouble than anywhere else. Authentication is not just about the end user logging in. It's about the agent itself having the right credentials to act on the user's behalf across multiple tools and systems. If the agent connects to a CRM, a calendar, a document store, and an internal API, each of those connections may have its own auth requirements. MCP (Model Context Protocol) is increasingly the standard way to handle these tool connections cleanly, but even with MCP in place, someone has to define who the agent can act as, under what permissions, and with what limits. This question has legal and security implications, not just technical ones and it rarely gets the attention it deserves until something goes wrong.</p><h2>4. What tools does it have and which ones read versus write?</h2><p>The tools are the agent's capabilities, that is the entire answer to what it can actually do. Which systems does it connect to, which information sources does it draw from, what problem is it solving with them? This is where MCP earns its place: well-documented, standardized tool definitions the model can call reliably. But listing capabilities is only half the question. The read/write distinction is where risk enters: reading from a database is low-stakes, while writing to production systems, sending emails, or updating customer records can cause real damage if the agent is manipulated or simply wrong. And prompt injection through untrusted content is a genuine attack surface. That's why guardrails belong at the tool level: tight permission scoping, confirmation steps before irreversible actions, hard limits the tools enforce regardless of what the model decides. Guardrails at the model level help; guardrails at the tool level hold. You always need to map out what is the catastrophic failure maximum impact and be content with that.</p><h2>5. What does the user interface look like?</h2><p>Not all agents live behind a chat window. Some are invoked from a command line. Some listen for voice. Some are triggered by an event like an incoming email, a form submission, a calendar alert without direct user interaction at all. The interface shapes the entire interaction model: how the user provides context, how the agent communicates back, and how the user intervenes mid-task. It's also a significant workload for the project to create the user interface which again is more of a traditional software development issue.</p><h2>6. What does it need to remember?</h2><p>Memory in agents exists on a spectrum. At the simplest end, the agent remembers only the current session. More capable agents remember previous conversations. More capable still, they maintain persistent knowledge about the user like preferences, past decisions, ongoing work leading effectively to a long-term operational context. At the far end, some use cases require returning to a specific checkpoint in a past conversation and branching from there. Each step up the memory ladder adds complexity, storage requirements, and privacy considerations. We've found that clients almost always want more memory than they actually need. Defining what the agent must remember to do its job usually results in a simpler and more reliable system. Minimum for all agents is to at least have "write to guidelines" tooling. But then the next question immediately becomes: Should those guidelines be shared between users? That again depends on the case.</p><h2>7. How do you measure whether it's working?</h2><p>An agent without measurement is a black box. Quality in agents is harder to define than in traditional software. It always answers something so you're often evaluating judgment, not just correctness. Does the agent complete the task? Does it complete it accurately? Does it know when to stop and ask for help rather than proceeding with incomplete information? Before deploying, you need to define what "good" looks like: task completion rate, error rate, time-to-completion, user satisfaction, or some combination. Without this, you don't have a way to detect for example whether a model update upgrades or degrades performance. Something we've seen catch teams off guard more than once.</p><h2>8. How do you gather feedback and improve based on it?</h2><p>Measurement tells you how the agent is performing. Feedback tells you why it's failing and what to do about it. Feedback loops can be explicit like users rating outputs, flagging errors, correcting the agent's work or implicit, derived from behavioral signals like task abandonment or repeated retries. Both matter. A well-designed feedback loop turns every failure into signal. Over time it becomes one of the most valuable assets the system produces, and one of the clearest competitive advantages for teams that build it early. First iteration of an agent never is very good, but with a fast feedback loop the 5th to 10th iteration usually is!</p><h2>9. How does the agent improve over time?</h2><p>This is the question most teams skip when they're focused on shipping and the one we always make time to answer before a project closes. Agents can improve through periodic fine-tuning on accumulated feedback, through better tool definitions and documentation, through refined system prompts based on observed failure patterns, or through architectural changes as the problem becomes better understood. Some improvements happen automatically; most require deliberate work. An agent that isn't designed to improve is one that gets relatively worse as the business around it evolves.</p><h3>Finally</h3><p>These nine questions come from a year of building agents across different industries, team sizes, and integration environments. None of them have universal answers and that's the point. What they do is replace the vague ambition of "we want an AI agent" with a concrete design brief. The agent you end up building should match the problem you actually have, not a generic approximation of one. And that's only the agent itself. Re-designing the business process around it is a separate checklist entirely. More on that another time.</p><p>And to close where we started: an agent operates on a fundamentally different timescale than a chatbot. It may take ten minutes to complete a task that would take a human two hours. That gap is where the business value lives but capturing it requires building something genuinely designed to act, not just to respond.</p><p>---</p><p><em>Jonas Pomoell is the founder and lead AI business consultant at Des Train AI Oy, specializing in practical AI agent implementation for business. If you're working through these questions for your own use case, <a data-mce-href="https://destrain.ai/contact" href="https://destrain.ai/contact">we'd be glad to help</a>.</em></p><p><br></p><p><br></p><p><em>Illustration Photo by Laura Chouette on Unsplash.</em></p>

What a year of agent projects taught us to ask before building anything

<p>Many clients are experimenting and requesting AI agents that query their databases. "Just ask questions in plain English and get instant insights!", sounds simple enough.</p><p>SQL is language and LLMs speak if fluently. However, the truth is making agents reliably talk to your production data is far more complex than the demos suggest.</p><h2>The Three Hidden Challenges</h2><p>Simply put, here are three fundamental issues that make "talk to your data" much harder than it appears. Especially when you're dealing with real business systems, not demo databases.</p><h3>1. The Performance Trap: One Query, Many Paths</h3><p>Here's how database queries work in practical terms: there's usually more than one way to get the same answer. In a small demo database, who cares? But in your production environment with millions of records, one approach returns results in milliseconds while another locks up your system for minutes.</p><p>When you unleash an autonomous AI agent on your database, you're essentially rolling the dice. Will it choose the efficient path with proper indexing? Or will it construct a query that sends your database administrator into panic mode?</p><p>For example, imagine querying customer orders from the last month. The agent could:</p><ul><li>Use the indexed date field (fast)</li><li>Scan every single order and filter by date (disaster)</li><li>Join multiple tables unnecessarily (somewhere in between)</li></ul><p>Without careful engineering, you have no guarantee which approach your agent will take – today, tomorrow, or when the CEO is waiting for that critical report.</p><h3>2. The Variance Problem: When Consistency Matters</h3><p>Let's be honest about how businesses actually use data. You're not asking unique, creative questions every day. Most data queries are repetitive – and that's by design.</p><p>Think about your typical reporting workflow:</p><ul><li>3-5 core tables you query regularly</li><li>3-5 standard queries for each table</li><li>These 9-25 queries probably cover 90% of your daily needs</li></ul><p>Why does this matter? Because these queries feed into dashboards, reports, and downstream processes. Your finance team needs the monthly revenue calculated the same way every time. Your inventory system depends on consistent stock counts. You probably have a definition of a "active account" or "paying user" that needs to be calulated the same way every time.</p><p>But AI agents thrive on creativity. Ask "show me last month's sales" twice, and you might get two different interpretations. Was that calendar month or last 30 days? Gross or net? Including pending transactions or only completed? By category or total?</p><p>What seems like helpful flexibility becomes a liability when you need reliable, reproducible results.</p><h3>3. The Translation Challenge: Your Database Speaks Its Own Language</h3><p>This is the killer: the more complex and specialized your database, the worse AI agents perform on it.</p><p>Generic LLMs are trained on publicly available patterns. They understand common database structures and standard business terms. But your production database? It's full of company-specific logic that would baffle any outsider.</p><p>What does an AI agent do when:</p><ul><li>The field "STAT_CD" having 'K' as the second character means an order has shipped?</li><li>"Customer Type 3" actually means enterprise clients (for historical reasons nobody remembers)?</li><li>The "deleted" flag being 0 means active, but only in certain tables?</li></ul><p>Your database is essentially speaking a dialect that evolved over years of business decisions, quick fixes, and "we'll document it later" moments. An LLM trained on generic data has zero chance of understanding these nuances without extensive preparation.</p><h2>The Reality: Your data investments pay out</h2><p>Here's what this means for your business: implementing "talk to your data" isn't just dropping in an AI agent and calling it done.</p><p>You still need:</p><p><strong>Performance optimization</strong> – Guiding agents toward efficient query patterns<br><strong>Template engineering</strong> – Creating reliable patterns for your common queries<br><strong>Semantic mapping</strong> – Teaching the AI your specific business logic and terminology, with data catalogs and other best practice patterns<br><strong>Validation layers</strong> – Ensuring consistency in critical reporting scenarios</p><p>The promise of natural language data access is real, but the path requires the same thoughtful data engineering we've always needed, now combined with agentic AI expertise.</p><h2>Making It Work in Practice</h2><p>Companies need to approach this challenge with eyes wide open. The winners aren't trying to skip on the data work overnight. They're strategically deploying AI agents where they add value while maintaining control where consistency matters. They are building the necessary data layers, handbooks and dictionaries for the agents to utilize the data. You don't want to build black boxes or externalize your ontologies.</p><h2>The Path Forward</h2><p>The direction is clear: AI agents will revolutionize how we interact with data. But the journey requires more than just plugging in an LLM and hoping for the best.</p><p>Success comes from recognizing that your production data environment is complex, specialized, and mission-critical. It demands the same rigor in implementation that you'd apply to any core business system, plus new expertise in making AI agents work reliably within those constraints.</p><p>Ready to implement AI agents that actually understand your data? We've learned these lessons the hard way, so you don't have to. <a data-mce-href="https://destrain.ai/contact" href="https://destrain.ai/contact">Contact us</a> to discuss how to make natural language data access work in your real production environment.</p><p>---<br><em>Jonas Pomoell is the founder and lead AI business consultant at Des Train AI Oy, specializing in deploying transformative generative AI solutions. With deep expertise in AI-driven business automation and data systems, Jonas helps companies navigate the complexities of implementing AI agents in production environments.</em></p>

Common pitfalls of "Talk to your data"

<p>When I'm talking about AI agents, I rarely talk about replacing people. Instead I turn the conversation into how AI can complement the employees in their every day work.</p><p>The more autonomous AI becomes, the more critical human expertise becomes. Simply put, AI agents are brilliant executors, but they need expert direction to deliver real business value. This is especially true when talking about <a data-mce-href="https://destrain.ai/blog/what-is-an-ai-agent--the-definition" href="https://destrain.ai/blog/what-is-an-ai-agent--the-definition">real AI agents, not automated workflows</a>.</p><h2>Autonomous but not independent</h2><p>Yes, AI agents can do a lot of tasks. They can write reports, send emails, analyze and input data - even make plans and decisions. But the autonomity only drives value when complemented with experience and vision, rules which determine the success: the quality of human understanding guiding them.</p><p>An AI agent without domain expertise is like an orchestra without a conductor. Full of potential, but without someone to shape the performance. It might play in tune for a while, but sooner or later it will lose direction. The expert brings the strategic insight, specialized understanding, and attention to detail that turn scattered notes into a symphony.</p><p>AI agents operate autonomously, yet their independence ends where context begins. They move on their own, but only within the boundaries of human insight and strategic intent. And remember, that if they're not delivering these values, <a data-mce-href="https://destrain.ai/blog/almost-everything-called-an--ai-agent--isn-t" href="https://destrain.ai/blog/almost-everything-called-an--ai-agent--isn-t">they might not be agents at all</a>.</p><h2><span style="font-family: 'Barlow Semi Condensed', sans-serif; font-size: 60px; font-weight: bold;" data-mce-style="font-family: 'Barlow Semi Condensed', sans-serif; font-size: 60px; font-weight: bold;">Why Experts Become More Valuable</span></h2><p>AI agents excel at execution, but they need three things only humans provide:</p><p><strong>Context and judgment.</strong> AI can process information, but understanding which information matters in your specific business context? That requires experience. Your industry knowledge, customer insights, and understanding of nuance. Those things transform AI from a tool into a strategic advantage.</p><p><strong>Quality control and direction.</strong> Every AI output needs expert validation. Not only because AI makes mistakes, but because business decisions require human judgment about risk, brand voice, and strategic alignment. The expert eye catches what matters and the envetual mistakes too, which are essential of LLMs to provide the content.</p><p><strong>Creative problem-solving.</strong> AI follows patterns, it formulates what has already been created. Experts break them. When your business needs innovation, competitive differentiation, or solutions to unprecedented challenges, that's where human creativity becomes irreplaceable. After all, humans have immensly wider context windows.</p><h2>What This Means for Your Business</h2><p>A recent Wharton study shows that shows that AI implementation winners, those that have the most measured impact, are using it FOR THE BORING STUFF. That means the backoffice tools, the repetative things in everyday work and bottlenecks that used to make some processes impossible. Doing more with the same effort and tools. </p><p>However, what they're not doing is running after the lastest shiny big thing of the month.</p><p>This has been evident already for a while. Already a year ago in Grand One AI Day in Helsinki 7.11.2024, Päivi Arffman-Koskinen, a Business strategist from Omnicom Media Group, noted in their presentation that they do not see decrease in campaing budgets when using AI. Instead they see 10-25 % increase in effectivenes of the campaing.</p><h2>Your Experts, Amplified</h2><p>At Des Train AI, we're experiencing firsthand how this partnership transforms businesses. We don't implement AI to reduce headcount, we deploy it to make your existing experts exponentially more effective. We want to study your processes and create narrow, effective and to the point solutions fast to gain the advantage.</p><p>Use our expertice in Python, AI agent frameworks, LLM models, MCP and the architectural landscape and help you transform your business.</p><p>The most successful AI implementations we've seen share one characteristic: they're led by companies that value their human expertise and use AI to supercharge it. Allow t<span style="font-family: inherit; font-weight: inherit;" data-mce-style="font-family: inherit; font-weight: inherit;">hem to execute at unprecedented speed and scale while maintaining their unique insights and judgment.</span></p><p>Ready to see how AI agents can amplify your team's expertise? <a data-mce-href="https://destrain.ai/contact" href="https://destrain.ai/contact">Reserve 30 mins from our calendar</a> to and we'll take a quick overview of your business together for free.</p><p>--<br><em>Jonas Pomoell is the founder and lead AI business consultant at Des Train AI Oy, specializing in deploying transformative generative AI solutions. With deep expertise in AI-driven business transformation, Jonas helps companies unlock productivity and growth through advanced, practical AI agent implementations.</em><br></p><p>(1) A business journal from the Wharton School of the University of Pennsylvania Accountable Acceleration: Gen AI Fast-Tracks Into the Enterprise, 12.11.2025, <a data-mce-href="https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report/" href="https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report/" target="_blank" rel="noopener">https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report/</a></p>

AI Agents Need Experts – Why People Matter More Than Ever

<p>According to Gartner, 95% of what's being promoted as "AI agents" aren't actually agents at all.</p><p>We're experiencing what Jordan Wilson from Everyday AI podcast brilliantly calls the "overagentification of everything." (E582)</p><p><em>If you're not totally sure about what makes an AI Agent read our previous post </em><a data-mce-href="https://destrain.ai/blog/what-is-an-ai-agent--the-definition" href="https://destrain.ai/blog/what-is-an-ai-agent--the-definition"><em>What Is an AI Agent? The Definition</em></a><br></p><h2>The great mislabeling</h2><p>Simply put, companies are slapping the word "agent" on anything that even touches an LLM. As Wilson points out: "Much of the market hype misuses the term 'agent'. Anything with LLM isn't agent, agentic or even meaningful AI."</p><p>But why does this matter to your business?</p><p>Because this confusion is creating a massive gap between what businesses think they're buying and what they're actually getting.</p><h2>What people are asking for</h2><p>Here's a fascinating insight from Daniel Whitenack, CEO of Prediction, from the Practical AI podcast (E352): "Every time I've actually asked people about what they want with their AI agent, it turns out what they want is either RAG chatbot or automated workflow."</p><blockquote><p>If I had asked people what they wanted, they would have said faster horses.</p></blockquote><p>Think about that. Many businesses can't even imagine yet what agents are capable of. They're missing out on these dynamic assistants and turning instead to incremental upgrades: slightly smarter chatbots accessing company knowledge or predefined automated workflows with a sprinkle of increased capabilities compared to what was done 10 years ago already.</p><p>They are valuable tools. They improve efficiency. They save time. A little bit.</p><p>Agent's transform whole processes.</p><h2>The hidden cost of confusion</h2><p>This mislabeling isn't harmless marketing fluff. When companies label their LLM wrappers "agents" their customers think that "this is it". At the same time they dillute the real possibilities of the actual dynamic agentic shift. They become disillusioned with AI's actual potential. They fall behind competitors who understand the difference.</p><p>What does this mean for your business? You might be settling for a chatbot when you could have a digital colleague. You might be buying into the 95% when you should be looking for the 5%.</p><h2>Why We're Writing This</h2><p>At Des Train AI, it's crucial that our customers know exactly what they're buying. We've had directors from large companies come to us stating that "Agents are nothing, they will never work". That smells like something a person disappointed with something sold to them as "agent". We can show they work, I'm running multiple agents as we speak.</p><p>When we say "AI agent," we mean systems that dynamically figure out how to solve problems using the existing tools and complementing the persons doing the work. Not because it sounds impressive, but because that's the technology that actually transforms businesses. Setting the right expectations upfront means you get the transformation you're investing in, not another disappointment dressed up in AI buzzwords.</p><h2>How to Spot the Real Thing</h2><p>Here's your quick test. Ask your vendor: "Does this system decide HOW to accomplish tasks, or does it follow predetermined steps?"</p><p>If they start talking about workflows, templates, or configurations you're probably looking at the 95%, not the 5%.</p><p>Real agents don't need you to spell out every step. They figure out how to achieve the outcome you want. That's the transformative difference.</p><h2>The Path Forward</h2><p>Our mission is to complement people’s work with AI agents that help amplify the businesses they build.</p><p>Why? Because when businesses understand what real AI agents can do: dynamically solving problems, taking ownership of processes, adapting their approach based on context, it changes everything.</p><p>The technology for genuine AI agents exists today. One of the challenges is cutting through the "overagentification" to find it.</p><p>Ready to join the 5% using real AI agents? <a data-mce-href="https://destrain.ai/contact" href="https://destrain.ai/contact">Contact</a> us to see the difference.</p>

Almost Everything Called an "AI Agent" Isn't

Open book with turning pages. Textbook in hard cover on table. Education and studying concept. Blue grey background with copy space for text

<p>An AI agent is a system where AI dynamically directs its own processes and tool usage, maintaining control over how it accomplishes tasks.</p><p>Anthropic's definition above captures it pretty well. And this is a question I get a lot.</p><p>That word "dynamically" is everything. It's the difference between following a recipe and being a chef. It separates the agent from chatbots and copilots by enabling autonomy and action.</p><h2>The Core Difference<br></h2><p><strong>Not an Agent:</strong> You ask a chatbot to summarize the quarterly sales and it prints them to your screen. When you ask it to send it to your collague, it apologises for not being able to do that.</p><p><strong>Not an Agent:</strong> You tell a agentic process the steps it needs to take in order to analyze quarterly sales data. It fetches them according to your instructions, does the defined analysis and sends them where you have defined. You ask it to compare it with previous quarter and it freezes.</p><p><strong>Real Agent:</strong> You ask an AI agent to analyze sales data. It decides which data sources to examine, what timeframes matter, which metrics to compare, identifies anomalies you didn't ask about, and determines the best way to present its findings. It asks you for more detailed information and digs deeper into insights that interest you. When you see something it missed, you ask it to improve and it will.</p><blockquote><p>You tell it what you want to achieve, not how to achieve it.</p></blockquote><p>One needs instructions. The other figures it out.</p><p>One is a one trick pony. The other solves varying problems using varying tools and data sources.</p><p>One gives out what it wants. The other complements the user's expertise.</p><p>That's it. That's what makes something an AI agent. You tell it what you want to achieve, not how to achieve it. (You need to give it enough guidance of course. And the tools, you need to hand in the tools. It's never that simple.)</p><h2>What This Looks Like in Practice<br></h2><p>Think of it like asking someone to get coffee. You don't need to say "walk to the kitchen, grab a mug from the cupboard, pour coffee in the mug..." You just say "bring me some coffee" and they figure out the steps. Add "two sugars", and they disgruntly return to the kitchen, but they know what it means.</p><p>That's exactly how AI agents work. When you tell an agent "analyse the status of this shipment," you don't spell out every step. The agent reasons what is an shipment, where to get it's data, what checks are relevant and what needs to be done for different statuses. It might ask you clarifying questions or need instructions to improve the end result - but that's why you're there, you're the expert afterall. You're not the expert in reading a spreadsheet. Instead your an expert on understanding what is relevant of the information on the spreadsheet. Agents put the user's expertice is in the focus.</p><p>What does this mean for your business? Increasingly flexible process supporting tools complementing employees exprtise. Using the same tools, spreadsheets, instructions and database as always. Internal and external data sources and tools. You're not programming workflows anymore. You're delegating outcomes.</p><h2>The Outcome</h2><p>Software that needs instructions saves time. Software that figures things out supercharges productivity.</p><p>At Des Train AI, we focus on genuine AI agents for real business processes. The kind that dynamically control their own processes to achieve your goals. The kind that actually figure things out. Using your existing tools and infrastructure. Complementing your team's expertise.</p><p>Is your business ready for AI that doesn't just assist but actually does?</p><p>Ready to see what real AI agents can accomplish? <a data-mce-href="https://destrain.ai/contact" href="https://destrain.ai/contact">Contact us</a> to explore the possibilities.</p><p>---</p><p><em>Jonas Pomoell is the founder and AI business consultant at Des Train AI Oy, specializing in deploying transformative generative AI solutions. With deep expertise in AI-driven business automation, Jonas helps companies unlock productivity and growth through advanced, practical AI agent implementations.</em><br>&nbsp;</p><p><br></p><p><br></p>

What Is an AI Agent? The Definition

<p>AI agents are revolutionizing busines processes and supercharging teams in their daily work. They don't only assist but actually get things done by reasoning and performing actions using the same everyday tools people are using in their work. For this they need the USB-C of integrations, the Model Context Protocol or MCP. [Read more on that from our previous blog on <a data-mce-href="https://destrain.ai/blog/how-do-ai-agents-actually-do-things" href="https://destrain.ai/blog/how-do-ai-agents-actually-do-things" title="How do AI Agents actually do things?" data-mce-selected="inline-boundary">How do AI Agents actually do things?]﻿</a></p><p>This Model Context Protocol revolution is changing everything about business tool integration. Tools that were never on your "integration radar" are suddenly becoming critical for AI agent workflows. Error tracking systems, project management platforms, communication tools and services you've always accessed manually are now expected to work seamlessly with AI agents.</p><p>But here's what might surprise businesses starting to deploy AI agents: many modern, everyday tools still lack the comprehensive programmatic access that agents prefer. No API. No CLI. No filesystem access. And certainly no MCP integration. It's not a blocker but a speed bump.</p><p>This isn't just about what tools you use anymore. It's about a fundamental shift and it's happening on two fronts:</p><ol><li>are your everyday tools ready for the MCP world</li><li>are the services you provide prepared for customers who expect MCP compatibility?</li></ol><h2>The Two-Sided MCP Reality</h2><p>This transformation is hitting businesses from both directions:</p><p><strong>As a tool consumer: </strong>Are your everyday SaaS business tools easily integratable with AI agents? Can your agents actually work with your error tracking, project management, analytics, and communication platforms? Or will you need complex automation workarounds for tools that should offer direct access?</p><p><strong>As a service provider:</strong> Are your services ready for customers who expect MCP integration? When your clients start deploying AI agents, will your platform be the one that "just works" or the one that requires expensive workarounds?</p><p>The businesses recognizing both sides of this shift are the ones preparing for competitive advantage.</p><h2>The Great Tool Migration</h2><p>Take Google Firebase Crashlytics – used to manage millions of mobile apps daily, yet lacking API and CLI access for the basic operations that AI agents need to perform. And that is just one of the tools that feel thoroughly modern still operate with limited programmatic access. This pattern might be more common than many businesses initially realize.</p><p>This is creating rapid market pressure, especially for tools where switching costs are low and alternatives exist.</p><p>We're entering a period where integration capability will drive tool selection like never before. In SaaS categories where options are plentyful and switching costs are reasonable – error tracking, project management, analytics platforms, communication tools – the pressure to offer proper programmatic access is intensifying rapidly.</p><p>Why? Because when your competitor's AI agents can seamlessly work with their entire tool stack while yours require manual intervention or complex automation, the productivity gap becomes unmistakable.</p><p>At Des Train AI, we're expecting this pattern to accelerate. Companies evaluating new tools now ask about MCP compatibility upfront. Existing tools without proper API access are getting replaced, especially in categories where migration is straightforward.</p><h2>If switching is easy it should be done</h2><p>Everything can be integrated with AI agents – that's not the question. The question is efficiency and competitive timing.</p><p>When switching costs are high (legacy enterprise systems, specialized industry software, deeply embedded workflows), building computer use automation for MCP use makes perfect sense. These integrations are worth the complexity.</p><p>But when switching costs are low? When modern alternatives with native MCP support exist? When the tool is used frequently by agents? The calculation changes completely.</p><p>This is should be rapid change in easily-switchable tool categories. And <span style="font-family: inherit; font-weight: inherit;" data-mce-style="font-family: inherit; font-weight: inherit;">the tools that don't adapt quickly are losing customers to competitors who understand the new integration reality.</span></p><p><span style="font-family: inherit; font-weight: inherit;" data-mce-style="font-family: inherit; font-weight: inherit;">And especially if you provide any tools, services or interfaces online including portals and websites - even for consumers. For a service provide</span><span style="font-family: inherit; font-size: 16px; font-weight: inherit;" data-mce-style="font-family: inherit; font-size: 16px; font-weight: inherit;">&nbsp;the shift represents both risk and massive opportunity.</span></p><p><strong>The risk:</strong> Your internal processes, customers and partners will start expecting AI agent compatibility. If your platform requires complex workarounds while competitors offer direct integration, you'll lose to "the one that just works with our agents."</p><p><strong>The opportunity:</strong> Being MCP-ready positions you as the forward-thinking choice. When prospects are evaluating tools for their AI-enabled workflows, comprehensive integration support becomes a decisive factor.</p><p>Companies doing their MCP implementation now will have competitive advantages over those waiting to see how the market develops.</p><h2>The change is silent</h2><p>This isn't a distant future consideration. AI agents are being deployed today, and their integration requirements are driving tool selection decisions right now.</p><p>The question isn't whether MCP compatibility will matter. It's whether you make the right choices right now, both internally and with the broad spectrum of services you provide externally.</p><p>The AI agent revolution is incremental and is happening within the everyday business processes. There will be no press releases of the agent deployments from your competitors. You'll know when you see them pass you due to their increased efficiency and quality it's already too late. But what if you could be the one on the driver's seat?</p><p>Ready to evaluate both sides of your MCP readiness? The companies moving fastest are auditing their tool integration gaps while simultaneously planning their service integration capabilities. <a data-mce-href="https://destrain.ai/contact" href="https://destrain.ai/contact" title="Contact us">Contact us</a> to develop a strategy that addresses both your tool consumption and service provision in the MCP-enabled future.</p><p>Interested to discuss more? <a data-mce-href="https://destrain.ai/contact" href="https://destrain.ai/contact" title="Contact">Book a meeting!</a></p><p>--- </p><p>Jonas Pomoell is the founder and lead AI business consultant at Des Train AI Oy, specializing in deploying transformative generative AI solutions. With deep expertise in Model Context Protocol implementation and business tool integration strategy, Jonas helps companies navigate both the consumption and provision sides of the AI agent integration revolution.</p><p>[The cover image taken 19.8.2025, real life questionnaire for paying users from widely used tranlsation and localization platform <a data-mce-href="https://phrase.com/" href="https://phrase.com/" title="Phrase">Phrase</a>]</p>

Blog

What a year of agent projects taught us to ask before bui...

Common pitfalls of "Talk to your data"

AI Agents Need Experts – Why People Matter More Than Ever

Almost Everything Called an "AI Agent" Isn't

What Is an AI Agent? The Definition

Your tools need AI access. Your customers expect it too.