From use case to production
AI is changing business fast and the businesses acting now will have the advantage
AI agents can now review complex material, work across systems and carry tasks through several steps. We implement them, put them into production and look after them once they are running.
Example deployments
What AI agents can now take on
Review tenders, quotes and contracts
Tenders, quotes, contracts, briefs, submissions and supporting documents can be reviewed together rather than one file at a time.
- Project brief
- Tender documents
- Subcontractor quotes
- Standard terms
- Capability information
- Read the full pack
- Compare submissions against the brief and each other
- Find exclusions, qualifications and inconsistencies
- Apply relevant company information
- Prepare a structured first review
The commercial position and the response
Keep multi-step work moving
Work often stalls because somebody has to check an inbox, find a document, update a system, contact another person and remember the next action. An agent can carry that chain forward across systems and escalate when a decision is needed.
- Inbound requests
- CRM
- ERP
- Document stores
- Understand the request
- Retrieve the relevant information
- Update records
- Trigger the next action
- Follow up when something is missing
Exceptions and consequential actions
Maintain a current view of a project
Project information sits across email, Teams, project systems, cost reports and documents. An agent can bring those sources together and maintain a current view of progress, risks, actions and commercial position.
- Teams channels
- Project email
- Reports
- Project systems
- Cost information
- Bring together progress and issues
- Track decisions and outstanding actions
- Read commercial signals
- Prepare a current operating view
Priorities, intervention and project direction
Track obligations and deadlines
An agent can watch contractual obligations, certificates, supplier changes, correspondence and due dates, then bring changes or missing information to the relevant person.
- Obligation registers
- Correspondence
- Certificates
- Contractor information
- Due dates
- Monitor what is due
- Request missing evidence
- Follow up the parties involved
- Check returned documents
- Assemble the material for approval
Formal approval or lodgement
Review work that was never economical before
Lost bids, previous claims, scope changes, customer behaviour and recurring delivery problems can sit across thousands of files. The cost of reading and comparing that volume of information has changed, making analysis possible that was previously too expensive or time-consuming.
- Previous tenders
- Pricing
- Scope decisions
- Outcomes
- Historical records
- Compare what won with what lost
- Find where commercial decisions changed outcomes
- Identify repeating patterns
- Prepare a briefing before the next bid
How to change the next bid or the broader strategy
Use several specialist agents together
One agent does not have to do everything. A deployment can combine specialist agents for research, retrieval, analysis, checking, drafting and action around one piece of business work.
- The request
- Company information
- Previous work
- Pricing inputs
- External sources
- Gather relevant context
- Retrieve company information
- Check requirements
- Prepare a draft
- Pass work between specialist agents
Final review and release
How we work
From the first idea to an agent in production
There are two ways to begin.
I have a use case
Bring us the use case. We scope the work, design the agent, implement it in your environment and can manage it once it is running.
Start with your use case →Help me work out where to start
We begin with the part of the business you want to examine and look at the work, information, systems and decision points involved.
Work out where to start →Both routes then follow the same delivery stages
Understand
We establish how the work happens now, who is involved, what information it needs, which systems it uses and where work stops, waits or repeats.
Choose the first implementation
We rank the options by commercial value, effort and risk, then agree the first implementation with you.
The answer may be an agent, Copilot Studio, conventional automation, an existing application or a simpler system change.
Design
We define the platform, model, information access, integrations, controls, ownership, support and measures of success before the build starts.
- The information the agent needs
- What it can access
- The systems it connects to
- Where a person approves
- How its output will be evaluated
- Who owns it
- How it will be supported
Implement and test
We build the agent, connect it to the working environment and test it against real examples of the work.
We roll it out with the people who will use it, so they know when to rely on the agent and when to hand to a person.
Manage and improve
We monitor performance, errors, platform changes and whether the agent is delivering the expected result, then make agreed improvements.
Production design
What an AI agent needs in production
A demo shows what a model can do. Before an agent becomes part of day-to-day work, the following needs to be in place.
Ownership
Every agent needs somebody accountable for why it exists, how it is used and whether it is still providing value.
Information
Approved information sources are identified, structured where necessary and kept current.
Identity and access
We give the agent only the access its role needs and check those boundaries before it is introduced.
Integration
Connections into the systems, APIs and workflows the agent needs to perform the work.
Human judgement
Approval and escalation points wherever money, risk, reputation or consequential decisions are involved.
Evaluation
Expected outcomes and repeatable evaluation using real examples of the work, rather than impressions from a demonstration.
Monitoring and cost
We keep performance, errors, model behaviour, usage and running cost visible after deployment.
Support
Models, APIs, permissions and business processes change. The agent needs support, controlled updates and retirement when it is no longer required.
Technology and commercial model
Choosing the platform and model
The platform depends on what the agent needs to do, where people will use it, which information it needs and the systems it must connect with.
Choose the work
Copilot Studio or Agent Builder
Agent Builder suits lighter agents based on Microsoft 365 information. Copilot Studio suits agents that need more workflow, actions and managed connections.
Built around your environment
We build around your own tenant, information and subscriptions wherever the architecture supports it.
Why Inlight IT
We already manage the systems AI agents depend on
An agent relies on identity, permissions, Microsoft 365, business information, integrations, networks, security and support. These are environments we already design and manage for our clients.
Three disciplines in one team
We combine knowledge of how the work operates, the Microsoft environment it runs in and the AI platforms used to build the agent.
Business operations
We follow the work through the business, including the decisions, information, systems and hand-offs involved.
Technology delivery
We work across Microsoft 365, Azure, Entra ID, cybersecurity, infrastructure, automation and managed environments. That is where the agent’s identity, access, integrations and support operate.
AI implementation
We implement agents across Microsoft Foundry, Copilot Studio, Agent Builder, OpenAI, Claude and the surrounding integration and automation stack.
Our team is already building agent-based solutions across Microsoft Foundry and Microsoft 365, including agents that work across company information, tenders and subcontractor quotes.
Ways to work with us
Project work or ongoing management
Project work
Design and implementation for a defined agent or piece of work, from scoping through to production.
Ongoing management
Monitoring, support, permissions, controlled changes and further development for agents that are already running.
Alongside internal IT
Additional AI architecture, integration and implementation capability while the internal team retains day-to-day ownership.
Common questions
Questions businesses ask before deployment
Do we need to have a use case already?
No. If you have a use case, we can move directly into scoping. If you do not, we can examine a part of the business and rank where agents could provide the greatest commercial value.
Can you manage agents we already run?
Yes. We can manage monitoring, support, permissions, platform changes and cost visibility for existing agents or Copilot environments. Support is raised and tracked in Microsoft Teams.
Do we need additional Microsoft licences?
Not necessarily. Some agents use existing Microsoft 365 services, some require Copilot or Copilot Studio licensing, and others use consumption-based Microsoft Foundry services. We identify the licensing and usage costs during design.
What is Microsoft Agent 365?
Agent 365 is Microsoft’s control plane for registering, governing and managing agents across an organisation. It covers agent inventory, identity, access, security and lifecycle management. Agent 365 has been generally available since 1 May 2026. When it forms part of the environment, its licensing and controls are included in the deployment design.
Who owns what you build?
You do. We design the solution around your environment and subscriptions where that architecture makes sense, and we avoid introducing an unnecessary proprietary layer that would keep you dependent on us.
What can the agent access?
Only what we grant it. Information sources, identities, permissions, tools and system connections are defined as part of the deployment. Microsoft 365 Copilot works within existing permissions, while custom agents are given access according to their specific role.
Can an agent take action in our systems?
Yes, where the architecture and controls support it. Agents can use tools, APIs and workflows to move work forward, with a person approving consequential actions.
What happens when an agent makes a mistake?
Mistakes are handled through bounded access, human approval, logging, monitoring and repeatable evaluation. The controls depend on what the agent can access and the consequences of the actions it can take.
How are implementation fees and running costs handled?
They are kept separate. Our fees cover the agreed design, implementation and support work. Platform usage, model consumption and Microsoft licensing are identified separately and can be monitored once the agent is running.
Are you limited to Microsoft models?
No. Microsoft Foundry is our lead production platform in Microsoft environments, but the model is selected for the work. OpenAI, Claude and other supported models can form part of the solution.
Work with Inlight IT
Talk to us about AI implementation
Bring us a use case or a part of the business you want to examine. We can scope the agent, implement it in your environment and look after it once it is running.
Thanks, we have it
We will contact you to discuss the use case or process and what the next step could involve.