--- title: "Installing a Way of Working with AI — Built to Adapt — Thoughtful Robots" description: "For AI to become part of how an organisation works, the capabilities behind it have to be put in place, given the right boundaries, and kept working as things change." source: "https://thoughtfulrobots.ai/articles/installing-a-way-of-working-with-ai-built-to-adapt" --- # Installing a way of working with AI — *built to adapt* For AI to become part of how an organisation works, the capabilities behind it have to be put in place, given the right boundaries, and kept working as things change. This article takes a closer look at the * *Installation & Maintenance* part of our approach to organisational transformation with AI.* Knowing where AI belongs is an important first step. Building the skills and understanding to work with it is another. But for AI to become part of how an organisation actually works, the capabilities that support these new ways of working need to be **put in place and kept working as things change**. Without this step, AI can remain largely an individual capability. People may still have to find the right information, copy context into AI, recreate prompts, move between systems and manually carry out the next step. **We are not installing AI components for their own sake. We are installing a way of working — and building it so that it can adapt.** For us at **Thoughtful Robots**, that work is practical. It includes helping organisations choose the right tools and vendors, setting up AI environments such as Claude or ChatGPT, connecting them to systems such as Freshdesk or Zoho CRM, installing reusable Agent Skills for particular workflows, configuring what AI can access and do, and putting evaluations in place so the capability can be tested as models, systems and ways of working change. We help organisations **put these capabilities in place, integrate them around the work people actually do, and maintain them as both the organisation and the underlying AI technology evolve.** ## Choosing the Right Tools and Vendors Once an organisation has identified where AI could help, one of the next questions is **what technology it actually needs**. The AI market is evolving quickly. Organisations may have to choose between different AI platforms, models, agent environments, knowledge and retrieval tools, integration products, security solutions and specialised AI services. At **Thoughtful Robots**, we can work alongside organisations to understand their requirements, **evaluate the available tools, services and vendors, compare their capabilities and trade-offs, and help decide what is appropriate to purchase and adopt**. The choice should be grounded in the work: the workflows being supported, existing systems, data and security requirements, level of control, cost, and how the capability will need to evolve. **The objective is not to recommend technology because it is new or popular. It is to help the organisation choose what actually fits the way it needs to work.** Once those choices are made, the next question becomes much more practical: **what actually needs to be put in place?** ## What Are We Actually Installing? Once the right tools and services have been chosen, they still need to be **configured and brought together around the work people actually do**. Consider a customer support team. A support executive receives a customer issue. They might currently search **Freshdesk** for the case history, look through product documentation, check the customer's details in **Zoho CRM**, bring the relevant information into an AI conversation, ask for help, return to the support system and then carry out the next steps themselves. The individual tools may already exist. **The missing piece is bringing them together into a working capability.** Thoughtful Robots might configure **Claude or ChatGPT** as the team's AI environment, connect it to **Freshdesk, Zoho CRM and approved product documentation**, and install a reusable **support-resolution Agent Skill** that captures the organisation's approved process for handling common support issues. The AI could then be given approved tools to retrieve customer information, consult the relevant product knowledge, prepare a response and update a support case. Permissions could allow it to perform some actions while requiring employee approval for others — for example, before issuing a refund or making a consequential change to a customer's account. We could also put in place a set of **representative support cases as evaluations**, so that when the model, skill or integration changes, the workflow can be tested again. **Together, that is the installed capability.** The technology matters because it makes a different way of working possible: instead of the employee repeatedly carrying information between AI and company systems, the required knowledge, connections, instructions and tools are already available within the workflow. ## Giving AI the Knowledge and Connections It Needs The support example also helps explain what sits underneath the experience. An AI model does not automatically have access to the organisation's current product documentation, customer history or support policies. Those sources need to be made available appropriately. For our support team, that might mean making approved product documentation available to AI and connecting the AI environment to **Freshdesk and Zoho CRM**, so authorised customer and case information can be retrieved when it is needed. Different mechanisms can make these connections possible. **Connectors** can connect AI applications to particular systems, while **MCP (Model Context Protocol)** provides a standard way for compatible AI applications to interact with external data, tools and services. The practical consequence matters more than the terminology: **the support executive no longer has to find and carry every piece of information to AI manually.** ## Giving AI the Ability to Help Do the Work Access to information is only part of the workflow. The support executive may also need to update a ticket, create a follow-up, retrieve additional customer information or prepare something for another team. This is where **tools and actions** become important. A tool gives AI a defined way to interact with another system. For example, we could give the AI a **Freshdesk tool** with specific actions such as **Get Ticket, Add Note and Update Ticket**, and a **Zoho CRM tool** with actions such as **Get Customer Record or Create Follow-up Task**. The AI can then choose the appropriate tool and action as it works through the support-resolution process — subject to the permissions and approval rules the organisation puts in place. **Giving AI knowledge helps it answer. Giving it access to tools allows it to help do the work.** The organisation may also have an established way of handling support issues. Rather than requiring every employee to explain that process to AI repeatedly, it can be captured as a reusable **Agent Skill**. For example, a support-resolution skill might instruct AI to gather the customer context, consult approved troubleshooting material, propose a resolution, identify whether escalation is required, prepare the ticket update and leave specified decisions with the support executive. **If people have to teach AI the same way of doing a task every time, that knowledge has not yet become an organisational capability.** ## Bringing It Together — with the Right Boundaries The knowledge, connections, tools and skills need somewhere to come together. Depending on the organisation and the workflow, that might be an AI application such as **ChatGPT or Claude**, or an agent environment such as **OpenClaw or Hermes**, where agents can use the available knowledge, skills and tools to carry out parts of a workflow. **An LLM provides the intelligence. The environment around it gives that intelligence what it needs to do useful work.** But connecting an AI environment to organisational systems does not mean giving it unrestricted access. Our support capability might be allowed to read the customer history, retrieve product information and prepare a response. Updating particular records might require approval. Issuing a refund might remain entirely with a person. Permissions, access controls and approval points therefore become part of what is configured. **Installing the way of working also means installing its boundaries.** At **Thoughtful Robots**, this means not only installing the appropriate environment, but configuring the connections, skills, tools, permissions and approval points around the particular workflow. ## What Exactly Are We Maintaining? Suppose our support capability has been running successfully for six months. Then **Freshdesk changes an API**. The support team introduces a new escalation process. Product documentation is updated. A permission policy changes. And a newer Claude model becomes available that appears better at handling complex support cases. The capability may still be running, but parts of the installed way of working now need attention. The **Freshdesk integration** may need updating. The product knowledge needs to reflect the latest documentation. The support Agent Skill needs a new version to incorporate the escalation process. Access controls may need changing. And the new model should be tested against the existing support evaluations before replacing the model already in use. This is what **maintenance** means in practice. It includes keeping knowledge current, integrations working, skills aligned with business processes, permissions appropriate, and models and dependencies up to date. It also means rerunning evaluations when something changes to check that the complete workflow still behaves as expected. **If AI becomes part of the workflow, evaluating AI becomes part of operating the workflow.** And maintenance is not only about fixing things that break. New models and capabilities will continue to appear. Part of maintaining the capability is deciding whether they genuinely improve the workflow, testing them, and upgrading deliberately when they do. **Maintenance is not simply keeping the technology running. It is keeping the installed way of working useful.** That is what **built to adapt** means. ## Leaving the Organisation in Control The final part is making sure the capability can live within the organisation. That means handing over more than access to an AI environment. It can include the **Agent Skills and their versions, integration documentation, permissions and approval design, evaluation suites, configurations, runbooks and upgrade procedures** that support the installed capability. At **Thoughtful Robots**, we help organisations install and configure these capabilities and can continue to support their maintenance and evolution as models, systems and organisational needs change. **Our goal is to leave the organisation with the understanding and capability to operate, maintain and evolve what has been put in place.** That is ultimately what turns an AI installation into an **organisational capability**. AI will keep changing. So will the organisations using it. The aim is to build ways of working that can **change with both**. Figuring out where AI belongs — in your product or how your organisation works? **Let's work through it together.** Explore our [workshop](/workshop). Figuring out where AI belongs — in your product or how your organisation works? **Let’s work through it together.** ### Learning to work *with* AI The practical skills, technical building blocks, role-specific understanding and judgement people need to use and shape AI in everyday work.