Ask a company what it wants from AI and the first answer may be broad: an agent, an automation platform, perhaps a generative AI feature. If we do stay with the conversation for a little longer, however, the problems they are looking to solve usually become much more specific. Customer emails are taking too long to sort. Employees cannot find the information they need. Staff are copying data between systems that were never designed to speak to one another.
Those everyday frustrations have become a recurring theme in Slash’s client and sales conversations. Companies are interested in AI, but the technology itself is rarely the end goal.
Most companies don’t come to us because they want AI as a technology. They usually have a process that is too manual, too slow, or too dependent on people moving information between systems.
– Razak
The discussion may start with AI agents, automation, or generative AI. Once the workflow becomes clear, the conversation usually shifts toward a more practical question:
Where can AI create measurable business value?
Read More: The state of AI in 2025: Agents, Innovation, and Transformation
What are businesses actually asking AI to do in 2026?
Most companies want practical AI applications that improve existing workflows rather than completely autonomous systems.
Fully autonomous systems attract attention, but client requests tend to centre on familiar operational problems:
- Reducing repetitive manual work
- Searching internal company knowledge
- Processing documents and emails
- Supporting customer service teams
- Qualifying sales enquiries
- Automating internal operations
- Connecting information across systems
- Helping employees make faster decisions
A company may approach Slash with plans to build an AI agent, only to discover during the mapping process that the real requirement is often more specific. The system might need to understand an incoming request, retrieve relevant information and recommend an action, with uncertain cases passed to an employee. That is a more concrete brief, and one that can be evaluated against the way the business already operates.
The initial request is often ‘we want an AI agent.’ Once we understand the workflow, the real opportunity is usually much more specific. That’s where the business case becomes clearer.
– Razak
Why are companies moving away from generic AI experiments?
Businesses increasingly want AI projects tied to a specific workflow, user group, and measurable outcome.
The early phase of generative AI gave many companies a reason to test what the technology could do. These days, the questions have become more exacting. A compelling demonstration is no longer enough; companies want to know whether a system can work with their data, fit into an existing operation, and produce a result worth paying for.
That scrutiny tends to surface a practical set of questions:
- How much time could this save?
- Which process should we automate first?
- Can AI work with our existing systems?
- How accurate does it need to be?
- Where should employees remain involved?
- What happens when AI is uncertain?
- How will we measure ROI?
This changes the conversation from an AI capability discussion into a business case with clearer costs, responsibilities, and measures of success.
A demo can look impressive, but the next question is always whether it can work inside the business and whether the value justifies implementing it.
– Razak
Read More: 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025
What pain points are businesses trying to solve with AI?
Manual information handling remains one of the clearest AI opportunities.
Across many organisations, employees still spend a considerable part of the day reading, searching, categorising and summarising information, then transferring it into another system so somebody else can continue the work.
The process often unfolds in a familiar way:
- Information enters through an email, document, CRM record, or customer request.
- An employee interprets the information.
- The employee decides what should happen next.
- Data moves into another system.
- Another person continues the process.
AI can assist at the point where information first needs to be understood. In an enquiry workflow, for example, it could identify the customer’s intent, retrieve account details, search internal documentation and suggest an action. Complicated or uncertain cases would still go to an employee.
The value comes from reducing the effort surrounding an existing task. The underlying work remains recognisable, but fewer hours are lost to searching, sorting and moving information by hand.
Are businesses actually asking for AI agents?
Yes, but the business requirement matters more than whether the final solution qualifies as an AI agent.
While the phrase “AI agent” can indeed describe a wide range of systems, the more important issue is whether the process genuinely needs an agent’s ability to select tools, evaluate context, and determine its next action.
We try not to start with the assumption that an agent is the answer. Sometimes a simpler AI-enabled workflow solves the problem better and with less complexity.
– Razak
The useful question is not:
Can we build an AI agent?
It is:
Does this process need an AI agent to achieve the business outcome?
What do businesses misunderstand about AI automation?
A common misconception is that successful AI automation requires removing people from the process completely.
In practice, AI systems can operate at several levels of autonomy. Depending on the task, they may:
- Provide information to an employee
- Recommend the next action
- Draft an output for approval
- Execute low-risk actions automatically
- Escalate uncertain cases
- Complete defined multi-step processes
Full autonomy is only one option.
Full autonomy sits at one end of that range. For many businesses, human review is a more practical place to begin because it allows the company to save time while retaining control over consequential decisions. The level of oversight can then change as the system proves itself and the organisation gains a clearer view of the risks.
The goal isn’t always to remove the human. In many cases, the better outcome is helping someone make a decision faster or removing repetitive work around that decision.
– Razak
Read More: The Human Side of AI Adoption: Lessons From the Field
Why does integration become such a big part of AI projects?
AI creates limited operational value when it cannot access the information and systems required to complete real business tasks.
Useful business AI may need to connect with:
- CRM platforms
- Internal databases
- Knowledge bases
- Customer support systems
- ERP software
- Document repositories
- Communication tools
- Custom APIs
Answering “What is our refund policy?” requires access to documentation.
Answering “Is this customer eligible for a refund, and what should happen next?” may require policy information, order history, CRM data, payment information, and business rules.
At that point, the quality of the model is only part of the project. Data access, system connections and permissions determine what the AI can see and what it is allowed to do. Integration therefore becomes central to whether the product works beyond a demonstration.
The model is only one part of the project. The real work often starts when you ask what data the AI needs, which systems it must connect to, and what it should actually be allowed to do.
– Razak
What concerns do companies have before adopting AI?
Companies commonly focus on ROI, reliability, data security, integration complexity, and operational control.
A technically capable system can still fail in practice. Employees may not trust it, the integrations may prove too difficult to maintain or the economic benefit may never become clear. These concerns are part of the product design, rather than questions to be postponed until after a pilot.
What separates a useful AI project from AI hype?
A strong AI use case connects a real business problem with accessible data, repeatable demand, measurable outcomes, and an appropriate level of automation.
At Slash, the early assessment usually considers questions such as:
- What problem does the business need to solve?
- Who performs the task today?
- Where does the current process create friction?
- What information and systems does the solution require?
- How will the business measure improvement?
One of the most useful things we can do early is reduce the scope. Instead of asking how AI can transform the whole company, we identify one process where the value can actually be tested.
– Razak
Narrowing the scope gives the project a defined workflow, an owner, and a measure of success. It also gives the business a chance to learn how the technology behaves in its own environment before extending it into more processes.
What are businesses ultimately looking for from AI?
Most businesses are looking for leverage: they want teams to complete more valuable work without increasing operational complexity at the same rate.
That can mean:
- Faster customer responses
- Less manual document review
- Better access to internal knowledge
- Fewer manual handovers
- Faster sales qualification
- More consistent operational decisions
The exact implementation will differ from business to business, but the expectation is remarkably consistent: AI should improve how work gets done rather than becoming another technology layer.
What has Slash learned from AI client conversations?
The strongest AI opportunities usually emerge when the conversation starts with the workflow rather than the technology.
Companies do not necessarily need the most autonomous AI system.
They need a solution that addresses a meaningful problem, integrates with existing operations, handles uncertainty appropriately, and creates measurable value.
Companies do not always need the most autonomous system available. They need one that solves a meaningful problem, fits into existing operations, responds appropriately when information is unclear and produces value the business can measure.
For Slash, this means tracing where people spend their time, where information becomes difficult to process, and where a workflow begins to break down. The result may be an agent capable of taking several actions, an AI feature added to an established workflow, or conventional automation with no AI at all. The business case determines which approach earns its place.
