# Pre-Built vs. Custom AI Agents: What IBM and Microsoft's Launch Means for You

> IBM and Microsoft now offer 6,000+ pre-built AI agents. Learn when to buy pre-built, when to build custom, and what new research says about scaling.

- URL: https://incube.club/blog/pre-built-vs-custom-ai-agents
- Published: 2026-11-18
- Updated: 2026-11-18
- Author: Incube Team (Incube, https://incube.club)
- Category: Comparisons
- Tags: AI agents, Agentic AI, Enterprise AI, AI governance, Build vs. buy, AI infrastructure

**Quick answer:** Pre-built AI agents suit standard, high-volume tasks with light integration, such as meeting notes or password resets. Custom AI agents suit workflows that encode your competitive edge, regulatory rules or deep system integration. Most organisations need both: buy for generic work, then blend or build for core processes, with human approval, audit trails and a pause button from day one.

## Key takeaways

- **IBM and Microsoft's Imagine Enterprise** launched on November 10, 2026 with more than 6,000 pre-built AI agents and assets, delivered with forward-deployed engineers.
- **Pre-built agents fit generic work**; custom agents fit workflows that carry your competitive edge, compliance rules or deep integrations.
- **Governance is the bottleneck, not the model**: 75% of Chief AI Officers in IBM's November 2026 study say governance, risk or compliance delays AI-driven action.
- **Earned autonomy works**: let agents propose, let humans approve, and widen the agent's remit as it proves itself.
- **Decide workflow by workflow** using two questions: how standard is the process, and how deep must the integration go?

On November 10, 2026, IBM and Microsoft announced a platform that ships with more than 6,000 pre-built AI agents and assets. Six days later, Genpact launched its own platform for building and governing agents in enterprise operations. If you run operations, the message from the big vendors is loud and clear: why build an agent when you can pick one off the shelf?

It's a fair question. At Incube we design and build specialised agents for operations teams, so it's one we think about every day. The honest answer is slightly annoying: it depends on which part of your business you're automating. This guide explains what was actually announced, what the newest research says about scaling agents, and how to decide between pre-built and custom AI agents without betting the budget on a hunch.

## What did IBM and Microsoft announce this month?

IBM and Microsoft announced **Imagine Enterprise**, a platform for building, running and governing AI agents that carry out industry-specific processes. According to the [joint IBM and Microsoft press release](https://newsroom.ibm.com/2026-11-10-ibm-and-microsoft-announce-a-new-ai-agent-platform-designed-for-industry-specific-transformation), IBM Consulting brings a fleet of more than 6,000 AI agents and assets. Microsoft supplies the AI, cloud, data and security foundation.

The pitch rests on three promises:

1. **Start from pre-built agents** already designed for your industry, then adapt them.
2. **Fit into what you already run**, connecting to existing data, processes and systems.
3. **Govern everything** with identity, observability and AI-compliance controls built in.

Under the hood, the stack is very Microsoft: Foundry, Copilot Studio and GitHub Copilot to coordinate agents, plus Fabric, Entra, Purview, Defender and Azure for data, identity, governance and security. IBM adds its Consulting Advantage tools for designing and governing tailored agents. The first industry frameworks cover financial services, insurance and retail. Manufacturing, energy and the public sector are named as possible next steps.

Notice the delivery model, though. The release says clients get the platform by engaging IBM Consulting to deploy and run it, or by co-innovating with them, with "forward-deployed engineering support" for qualifying engagements. Self-management options exist too. Read that twice. The world's biggest agent catalogue still arrives with engineers attached.

## What is Genpact Onyx, and why does it matter?

Genpact Onyx is an "agentic operations" platform for building, running and governing AI agents inside enterprise operations. [Genpact announced its availability on November 16, 2026](https://www.prnewswire.com/news-releases/genpact-launches-genpact-onyx-the-agentic-operations-platform-for-building-running-and-governing-ai-agents-in-enterprise-operations-302940973.html), after unveiling it at an event in New York on November 12.

Two ideas in that announcement are worth stealing, whatever platform you use:

- **Process intelligence comes first.** Genpact argues that agents should start with context on "how the work gets done", including the edge cases and exceptions that "rarely show up in a prompt library or a static process map."
- **Autonomy is earned.** Agents begin by proposing actions for a human expert to approve. They take on more of each decision only as they prove themselves.

Here's the interesting bit. Genpact says Onyx powers its own pre-built solutions, starting with an accounts payable suite, *and* the custom solutions it builds for clients. So even a vendor selling packaged agents is telling you, in its own FAQ, that some work needs a custom build.

## What is the difference between pre-built and custom AI agents?

A **pre-built AI agent** is a ready-made agent designed for a common task, such as summarising meetings or matching invoices, that you configure rather than design. A **custom AI agent** is designed around one organisation's own workflow, data, rules and approval chain, and is built to fit them.

Think of it like suits. Off the rack is quick, affordable and fine for most days. Made to measure costs more and takes longer, but it's the one you wear when the fit actually matters. Most people own both. Most companies will too.

| Factor | Pre-built agent | Custom agent |
|---|---|---|
| Time to first value | Usually faster | Usually slower |
| Fit to your process | Generic, you adapt to it | Shaped around how you work |
| Edge cases and exceptions | Common ones only | Your real ones, from your data |
| Integration depth | Standard connectors | Your systems, fields and approvals |
| Competitive advantage | Low: rivals can buy the same agent | Higher: encodes how *you* operate |
| Vendor lock-in risk | Tied to the platform's roadmap | Depends on how it's built |
| Good fit for | Standard, high-volume tasks | Core, regulated or differentiating work |

Neither column wins outright. The trick is knowing which work belongs where.

When we scope an agent for an ops team, the first thing we look at is the exceptions, not the happy path. Ask for last quarter's ten weirdest cases: the invoice with two currencies, the customer who replies to a ticket from a personal email, the order that ships from a site nobody remembers opening. If a pre-built agent handles those, buy it. If your team's eyes light up with stories, you're probably looking at custom work.

## What does the new research say about scaling AI agents?

The newest research says the agent itself is rarely the bottleneck. Governance, decision rights and discipline are. A day after Genpact's launch, IBM published [a study of more than 1,600 Chief AI Officers](https://newsroom.ibm.com/2026-11-17-ibm-study-chief-ai-officers-see-where-ai-creates-value,-but-few-control-investment-decisions). The IBM Institute for Business Value and Oxford Economics surveyed 1,667 AI leaders across 33 geographies and 20 industries between April and June 2026.

A few findings should land on every operations leader's desk:

- **48%** say AI agents already initiate and execute actions across workflows.
- **75%** say governance, risk or compliance delays AI-driven action.
- Only **61%** say AI decision rights are clearly defined and consistently applied.
- Only **18%** have full control over AI investment decisions.

The study also singles out a group of "high-impact" CAIOs. Their organisations report that 63% of AI projects deliver measurable business outcomes, compared with 43% for everyone else. They also scale successful initiatives enterprise-wide in eight months, versus 13.5 months for the rest.

![Bar chart comparing high-impact Chief AI Officers with all others on four AI results and practices, from IBM's November 2026 study](https://incube.club/blog/images/bar-chart-from-ibms-november-2026-chief-ai-ca2c2772f5.webp "Source: IBM Institute for Business Value Chief AI Officer study, published 17 November 2026. Chart by Incube.")

What separates them isn't a secret model. It's habits. High-impact CAIOs are nearly four times as likely to say they can quickly pause, adjust or redirect AI systems without disrupting operations. And 96% of them stop funding initiatives that fail to deliver, even with senior sponsorship, against 46% of the others. In other words, they treat agents like a portfolio, not a pet project.

An earlier IBM survey of 2,000 CIOs and CTOs, [reported by ITPro in June 2026](https://itpro.com/technology/artificial-intelligence/cios-and-ctos-are-making-high-stakes-decisions-with-incomplete-information-ibm-survey-reveals), points the same way. Organisations that build control directly into their AI systems deploy 16 times as many AI agents as those relying on manual governance. Governance, it turns out, is not the brake. It's the steering wheel.

## Why do so many agent projects stall or get cancelled?

Agent projects stall when they are bought for hype instead of a clear business case. Back in June 2025, Gartner predicted that more than 40% of agentic AI projects will be cancelled by the end of 2027, [as Reuters reported](https://www.thestar.com.my/tech/tech-news/2025/06/25/over-40-of-agentic-ai-projects-will-be-scrapped-by-2027-gartner-says), citing escalating costs and unclear business value. Gartner also warned about "agent washing": rebranding chatbots and assistants as agents. It estimated that only about 130 of the thousands of agentic AI vendors are real.

So a catalogue of 6,000 agents is impressive, but the number alone tells you nothing about whether *your* invoice exceptions get handled. Count outcomes, not agents. (If you want a refresher on what separates a real agent from a dressed-up chatbot, see our guide to [AI agents vs. chatbots vs. RPA](https://incube.club/blog/ai-agents-vs-chatbots-vs-rpa).)

## How do you decide between pre-built and custom AI agents?

Decide workflow by workflow, using two questions: how standard is this process across companies, and how deeply must the agent plug into your systems, data and approvals? Those two answers place each workflow in one of four boxes.

![Two-by-two decision map: buy pre-built, blend, adapt a template, or build custom AI agents, based on how standard the process is and how deep the integration goes](https://incube.club/blog/images/two-by-two-decision-map-for-pre-built-c9fbbff33f.webp "A simple way to sort your workflows. Framework by Incube.")

1. **Buy pre-built** when the process is generic and integration is light. Meeting notes and password resets don't need a bespoke brain.
2. **Blend** when the process is common but the integration is heavy. Invoice processing is similar everywhere, yet your ERP fields, approval limits and vendor quirks are not. Start from a proven pattern and customise the edges.
3. **Adapt a template** when the logic is yours but the plumbing is simple, such as lead scoring against your own qualification rules.
4. **Build custom** when the work encodes your competitive edge or your regulatory obligations, and touches many systems. That's where off-the-shelf agents struggle with the exceptions that matter most.

### A worked example: one finance team, three answers

Picture a hypothetical mid-sized distributor with a busy finance team. It has three workflows on the shortlist.

- **Expense receipt checks.** Every company does them roughly the same way, and the agent only needs to read receipts and flag policy breaches. That's a clear *buy*. Grab a pre-built agent and move on.
- **Supplier invoice processing.** The steps are textbook: capture, match against the purchase order, route for approval. But this team has three ERPs after two acquisitions, approval limits that vary by site, and a handful of suppliers who invoice in creative formats. That's a *blend*. Start from a proven pattern, like the one in our [invoice processing agent](https://incube.club/work/invoice-processing-agent) walkthrough, and build the edges around their systems.
- **Rebate and credit-note disputes.** These follow contract terms negotiated customer by customer, live across email, spreadsheets and the ERP, and directly affect margin. No catalogue agent knows those contracts. That's a *build*.

Same team, same month, three different answers. That's normal. The mistake is forcing one answer onto all three.

Before any agent goes live, pre-built or custom, these are the five questions we ask. Steal them:

- **Exceptions:** Can it show you how it handled your ten weirdest cases from last quarter?
- **Control:** Can you pause, adjust or roll it back without stopping the process? IBM's data says the high performers can.
- **Decision rights:** Is it written down what the agent may decide alone, and what needs a human?
- **Audit trail:** Can you trace every action to a reason? Genpact puts this front and centre, and so should you.
- **Exit:** If you switch platforms in two years, what do you keep: the prompts, the data, the evaluation tests?

For a framework on governance, the [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) is a solid, vendor-neutral starting point. It's voluntary, free and widely referenced.

## What do these launches mean for mid-sized organisations?

These launches mean the agent pattern is now mainstream, but the delivery model is the real lesson. Imagine Enterprise comes with forward-deployed engineers. Onyx comes with Genpact's domain experts who "own the outcome." Both vendors are effectively saying that the hard part isn't the agent. It's the fitting.

That's great news if you're a global bank with a consulting budget the size of a small moon. It's less useful if you're a 300-person company that needs one agent to tame the shared inbox by March.

The good news is that the principles scale down nicely:

- **Start with one high-volume, rules-heavy workflow.** Our [5-step task audit](https://incube.club/blog/how-to-choose-your-first-ai-agent) helps you pick it.
- **Use earned autonomy.** Let the agent propose, let a human approve, and widen its remit as the error rate drops.
- **Build the guardrails on day one.** Logging, approval thresholds and a big red pause button are not optional extras.
- **Measure business outcomes,** such as hours saved, cycle time and error rate, and be willing to stop what doesn't work.
- **Get your AI infrastructure right early.** Decide where models run, how data is accessed and who owns each agent before you have twenty of them.

If agents are new to your team, our plain-English explainer on [agentic automation](https://incube.club/blog/what-is-agentic-automation) is a good place to start.

## What AI infrastructure do you need before your second agent?

You need four things in place before the second agent arrives: a model and hosting decision, governed data access, named ownership, and an evaluation and monitoring loop. One agent can survive on goodwill and a shared spreadsheet. Five agents cannot. Here's the checklist we work through when we set up the foundations for an organisation:

| Layer | The question to answer | What "done" looks like |
|---|---|---|
| Models and hosting | Which models run where: cloud API, private cloud or on-premises? | A short approved list, with a reason for each |
| Data access | What can each agent read and write, and through which systems? | Least-privilege permissions, logged |
| Ownership | Who owns each agent's results, and who can switch it off? | A named owner and a documented pause procedure |
| Decision rights | What may the agent decide alone? | Written thresholds, with human approval above them |
| Evaluation | How do you know it still works next month? | A test set built from real cases, run on every change |
| Monitoring | What gets measured in production? | Accuracy, cycle time, cost and escalation rate on one dashboard |

That last row matters more than it looks. IBM's study found that the high performers use business outcomes to decide which AI initiatives get more funding. You can't do that without numbers. It's also why we design agents such as a [support resolution agent](https://incube.club/work/support-resolution-agent) or an [inbox triage agent](https://incube.club/work/inbox-triage-agent) around escalation paths and human review, not just the model.

## What should you ask your team this week?

Ask one question in your next ops meeting: "Which workflow eats the most hours, and what are its ten weirdest cases?" The answer tells you almost everything. If the weird cases are rare and boring, a pre-built agent will probably do. If they're frequent, expensive or tied to your own contracts and systems, you've found a workflow worth fitting properly.

Then put that workflow on the decision map above, run the five pre-launch questions, and check the infrastructure table before you sign anything. Whatever IBM, Microsoft or Genpact put in their catalogues next, those steps won't go out of date.

Fitting agents to messy, specific workflows, and setting up the infrastructure they run on, is the work we do at Incube. If you'd like a second pair of eyes on your shortlist, [we're happy to compare notes](https://incube.club/#contact).

## Frequently asked questions

### What is the difference between pre-built and custom AI agents?

A pre-built AI agent is a ready-made agent for a common task, such as summarising meetings or matching invoices, that you configure rather than design. A custom AI agent is designed around one organisation's own workflow, data, rules and approvals. Pre-built agents are usually faster to start; custom agents fit unusual processes, exceptions and deep integrations better.

### What is IBM and Microsoft's Imagine Enterprise?

Imagine Enterprise is an agentic platform announced by IBM and Microsoft on November 10, 2026. It helps enterprises build, run and govern AI agents for industry-specific processes. IBM Consulting contributes more than 6,000 pre-built agents and assets, while Microsoft provides the AI, cloud, data and security technology. Initial industries are financial services, insurance and retail.

### When should a company build a custom AI agent?

Build a custom AI agent when the workflow carries your competitive advantage or regulatory obligations, has many exceptions that generic agents miss, or must connect deeply to your own systems, fields and approval chains. Examples include underwriting exceptions, claims handled against your policy wording, and invoice processing tied to your ERP and approval limits.

### Why do agentic AI projects fail?

Agentic AI projects usually fail because of escalating costs, unclear business value or weak risk controls, according to a June 2025 Gartner prediction that over 40% will be cancelled by the end of 2027. IBM research from November 2026 adds that unclear decision rights and slow governance hold back many organisations that already run AI agents.

### What is earned autonomy for AI agents?

Earned autonomy is an approach where an AI agent starts by proposing actions for a human to approve, then takes on more decisions as it proves accurate and safe. It lets teams capture value early while keeping control. Genpact built the idea into its Onyx platform, and it works just as well for smaller organisations deploying their first agent.
