The Return of Best of Breed: How AI Is Rewriting the ERP Debate

AI is making best of breed attractive again, but the winners will still anchor on a unified core. The difference is that the edge around that core will be built, connected and run by AI instead of by large integration teams.

For years, I used to preach the benefits of having all of the data in one unified core. Of course, I also used to be nervous about having healthcare data "in the cloud." We see where that went. For thirty years, enterprise software has swung between two ideas. Best of breed says buy the best tool for each job and stitch them together. The unified suite says run the business on one platform with one data model, even if some modules are only good enough. Through the 2010s the suite won most HR and finance decisions, and platforms like Workday were built on that argument.

Now the pendulum is moving. When an analyst can describe an app in plain language and have a working tool by the afternoon, and an AI agent can map fields between two systems in minutes, the old cost of going best of breed starts to fall apart. The question for CIOs, CFOs and CHROs is not which side wins. It is where each approach belongs, and what AI agents do in each.

Why best of breed lost the first time

Best of breed did not lose because the tools were bad. It lost because the space between the tools was expensive and supporting that many applications was even more expensive.

The integration tax. Every new system meant point-to-point interfaces, middleware licenses and a team to babysit them. Ten systems could mean dozens of integrations, each one a future break point.

Data that never agreed. A worker existed in HR, payroll, finance and the LMS, each with its own ID and its own definition of "active." Reporting turned into reconciliation.

Upgrade pain. Each vendor released on its own schedule. One API change could break payroll the week before a close.

Security and audit sprawl. Separate role models in every tool made access reviews and SOX testing slow and error prone.

Vendor management overhead. More contracts, more renewals, more support desks and more finger pointing when something failed.

Unified suites answered all of this with one tenant, one security model and one release calendar. Buyers accepted weaker functionality in some modules in exchange for lower total cost and cleaner data. The answer was always "the weaker functionality area would get better over time."

How AI changes the math

AI attacks the exact costs that sank best of breed: building tools, connecting them and keeping them running.

Building is cheap. AI coding tools let a small team, or a business analyst, produce a focused app for one workflow in days. A scheduling tool for clinics, a commission calculator or a grant compliance tracker no longer needs a vendor or a six-month project.

Integration is faster. AI can read API documentation, propose field mappings, generate transformation logic and write the tests. Work that took an integration developer weeks can be drafted in hours and then reviewed.

Maintenance gets lighter. When a vendor changes an API, an AI assistant can detect the break, suggest the fix and open the change for approval. Release notes can be read and impact assessed automatically.

Data can be reconciled on the fly. Models are good at matching records that do not share a clean key, like the same supplier spelled three ways.

The user interface matters less. If people ask questions in chat instead of clicking through screens, the system behind the answer can be one tool or ten. The front door is the AI, not the app.

The result is a new kind of best of breed: small, purpose-built tools that a company owns or rents cheaply, wired together by AI instead of by expensive middleware projects. It is cool again because it is fast, tailored and no longer requires a large IT budget to try.

The case for the unified system

The unified platform's strongest argument in the AI era is that AI is only as good as the data and context it can trust.

One data model. When worker, position, cost center and ledger live in one object model, an AI answer about labor cost by department does not depend on five systems agreeing. Fewer joins, fewer hallucinations.

One security model. AI that can read everything is a risk. A single role-based security framework means the AI sees only what the person asking is allowed to see, enforced in one place.

Built-in audit and controls. Business process history, approvals and effective dating are already captured. That matters when regulators and auditors ask how a decision was made, including one an AI recommended.

Vendor AI investment. Suite vendors are embedding AI directly into the platform, grounded in the customer's own tenant data. Customers get that capability on the regular release cycle without building it.

Less to govern. Every AI-built app is also an app someone has to secure, test, document and own when its creator leaves. A suite keeps that list short.

The risk of AI-fueled best of breed is not cost. It is sprawl: dozens of quickly built tools with no owner, inconsistent data and unknown access. The unified system is the antidote.

Side by side

AI narrows the gap on speed and fit, but the unified system still leads on data trust, security and governance.

Factor	AI-enabled best of breed	Unified system
Time to new capability	Days to weeks; build or buy a focused tool	Tied to vendor roadmap and release cycle
Fit to unique processes	High; tools shaped to how you work	Good; configuration within platform limits
Integration effort	Lower than before; AI drafts mappings and tests	Minimal inside the suite
Data consistency	Depends on AI reconciliation and discipline	Native; one object model
Security and access	Many models to align and review	One role-based framework
Audit and compliance	Assembled across tools	Built into business processes
Upgrade risk	Many release calendars; AI helps detect breaks	One release calendar
Ownership burden	Grows with every tool built	Concentrated with the vendor
AI quality	Limited by fragmented context	Grounded in one trusted data set
Best for	Differentiating, fast-changing work	Core records: people, pay, money, compliance

Where AI agents fit in each scenario

Agents play a different role in each model: in best of breed they are the glue, in a unified system they are the workforce.

In a best of breed landscape: agents as the connective tissue

Integration agents watch for changes in one system and push them to others, replacing brittle scheduled interfaces.

Orchestration agents run a process that spans tools, such as onboarding a hire across HR, IT provisioning, payroll and learning.

Reconciliation agents compare records across systems every night, flag mismatches and propose fixes.

Monitoring agents read vendor release notes, test integrations and alert the team before a change breaks production.

A single conversational front end lets employees ask one agent instead of learning ten apps.

The catch: agents moving data between many systems need their own identity, permissions and audit trail. Without that, the agent layer becomes the new spaghetti. That is why you are seeing major software companies buy or build that agent management layer.

In a unified system: agents as digital workers inside the platform

Embedded agents handle work inside the suite, like drafting job requisitions, matching supplier invoices or answering policy questions from employees.

Process agents move transactions through approvals, check them against policy and route exceptions to people.

Insight agents explain variances, forecast headcount cost or surface flight risk using the full, consistent data set.

Configuration and testing agents help admins assess releases, build tests and validate data loads.

Because they inherit the platform's security and audit, these agents are easier to trust and govern. The limit is reach: they are strongest inside the suite and weaker at work that crosses into other systems.

The convergence

The two models meet in the middle. Suite vendors are opening their platforms so outside agents can act through governed APIs, and agent frameworks are adopting shared standards for how agents call tools and talk to each other. That makes it realistic to run a core platform's agents alongside specialist agents from other tools, coordinated under one set of rules.

The likely outcome: a unified core with an AI-built edge

The winning architecture is not either/or. It is a strong unified core for the system of record, surrounded by a flexible edge of AI-built tools and agents.

The core holds people, pay, money, security and compliance. It changes slowly on purpose. This is where data quality and audit matter most, and where a platform like Workday earns its keep.

The edge holds the work that makes a company different: a custom scheduling model, an industry-specific compliance check, a niche analytics tool. It changes fast and can be built or replaced cheaply.

The agent layer connects the two, reading from the core, acting through governed APIs and writing results back so the core stays the source of truth.

Best of breed comes back, but in a new shape. Instead of competing with the core for the system of record, the new breed of tools sits beside it and makes it more useful.

What leaders should do now

Treat the core as a foundation to protect and the edge as a place to experiment, with clear rules for both.

1. Define your core. Decide which records must live in one governed platform. For most organizations that is workers, compensation, payroll, the general ledger and security.

2. Clean the core data. AI on both sides depends on it. Fix worker, position and cost center data before adding agents.

3. Set rules for the edge. Every AI-built tool needs an owner, a data classification, a security review and a retirement plan.

4. Govern agents like employees. Give each agent its own identity, least-privilege access, logging and a human accountable for what it does.

5. Use AI on the plumbing first. Integration mapping, release testing and data validation are low-risk, high-payoff places to start.

6. Measure the total cost. Count the time to own, secure and support each new tool, not just the time to build it. This includes AI tokens and credits.

Best of breed is cool again because AI made it cheap to try. The organizations that win will be the ones that keep their core unified and let AI do the building at the edge.

Evaluating your ERP architecture for the AI era? Talk to AssistNow.