I hear versions of this question often. If models keep improving, will they absorb the orchestration layer? Will connecting tools become a feature everyone gets for free?
Those are real competitive questions. But “just middleware” overlooks both the history of enterprise software and what it takes to put AI to work.
First, middleware has produced substantial businesses.
MuleSoft connected enterprise applications and APIs. Salesforce acquired it for approximately $6.5 billion in 2018. Confluent built a data streaming business that reported $1.17 billion in 2025 revenue.
These businesses solved different problems, but they demonstrate how valuable it can be to make complex systems work together.
Being middleware does not doom a company. The harder question is whether it solves a durable problem that customers will pay to have someone else manage.
AI workflow automation takes on another layer of responsibility.
Imagine an AI handling a customer refund.
Understanding the email is one step. Completing the job requires checking the purchase, applying the refund policy, confirming permissions, requesting approval when necessary, issuing payment, and updating the account.
If the payment succeeds but the account update fails, the system needs to recover without issuing a second refund.
That requires context, state, rules, execution, and accountability across systems. Better models help, but model intelligence alone does not establish who can authorize a payment or whether it actually happened.
Middleware provides connectivity. Workflow automation coordinates the process. AI expands the range of inputs and decisions that process can handle.
A valuable product can combine all three.
We are already seeing businesses emerge across that spectrum.
n8n reported 10x revenue growth during 2025. Workato reported 35% ARR growth in January 2026, with more than 700 customers contributing over $100,000 ARR each. Sierra reported more than $150 million ARR in February 2026 for its customer service AI business.
The business models reflect that spectrum. n8n charges by workflow execution. Sierra offers outcome-based pricing. The closer a company gets to owning the result, the more directly it can connect its price to customer value, while taking on more delivery responsibility.
The risk of becoming a feature is real.
A product that connects a model to a few APIs faces pressure from model providers, application vendors, and existing automation platforms. Our investment thesis has to explain why customers will keep choosing an independent company as those alternatives improve.
That means looking beyond the demo to what becomes difficult in production: permissions, exceptions, recovery, changing business rules, and proving the work was completed correctly.
What we are looking for as investors.
1. A painful process with an identifiable budget. A customer who can explain what the problem costs today and who is responsible for fixing it.
2. Responsibility for meaningful work. The product completes a valuable process, or provides infrastructure essential to completing it reliably.
3. Depth that compounds with deployment. Reusable integrations, domain knowledge, evaluation data, and exception handling that make subsequent deployments faster and better.
4. A credible reason to remain independent. Customers need capabilities across models and systems, or specialized depth that a single platform is unlikely to prioritize.
5. Economics that improve as usage grows. Revenue growth needs to translate into attractive margins after inference, implementation, support, and human review.
6. A clear path from an initial use case to expansion. Solving one problem earns the trust to handle adjacent work, with evidence that customers want that expansion.
I think AI workflow automation can support enduring companies because enterprises need someone to take responsibility for turning intelligence into reliable action.
For us, the investment question is: what does this company make possible, how reliably does it deliver, and why does it become more valuable as models improve?
Email us deals@array.vc if you are solving this.
Follow us on X, LinkedIn, & subscribe to newsletter to stay updated on all things AI, latest launches, AI research, or general market commentary :)