AI is for Builders, not Operations

What we’re learning in 2026 is that even though AI agents are getting better and more capable, AI is not a cure-all for business. Instead, it offers new patterns for technology development, business growth, and strategy that do not replace the underlying business, but instead streamline it.

AI succeeds in Development/Strategy

Where AI succeeds is in Development and Strategy. AI is a builder’s dream:

• Building software tools
• Building custom reports/analysis
• Building plans/processes
• Building media content
• Building compliance/regulatory/grant submissions

AI has made builders of all tiers: experienced professionals to generalists to dabblers, all more capable and more productive. More stuff can be built for your business than ever before.

AI struggles in operations

Where AI is struggling to contribute much, and where classical software is still dominating is Operations:

• Financial management: Accounting/Finance/Banking
• Supply chain/parts tracking/logistics
• Sales/leads/account management
• Staffing/HR

AI may demo well in these environments, but it fails miserably in practice: the smartest, most powerful LLMs in the world make critical errors, run slowly, and cost huge sums of money — while legacy software systems still hold up remarkably well.

AI strategy for 2026: Put AI-empowered Builders inside Operations

The solution is to put builders inside your operations. The biggest job of 2026 is the “forward deployed engineer” — the builder who lives on-site right alongside the doers, building the future while living inside it.

No longer is software development an obscure, specialized skill reserved for the geeks who have to be segmented off into their own department, protected from the rest of the company by layers of management. Increasingly, it’s a skill for ambitious, business-minded builders who want to make an impact, and they’re willing to get their hands dirty out there in the field alongside regular employees, so they really understand what it takes to make the business better.

Case study: LIS workflows

I am currently working with a startup that is building a Laboratory Information System (LIS) for clinical medical laboratories, and we came into this thinking that AI would be integrated into many features of the system. Instead, we have found ourselves pulling out AI feature after AI feature, replacing them with deterministic systems, which are faster, more reliable, and more power-user friendly (even if with a steeper learning curve).

Operational workflows we keep pulling AI out of

• Voice-driven applications: We have slowly replaced each LLM-powered feature of our voice application with an equivalent deterministic ruleset — thousands of lines of word lists and grammar rules that are 20x faster and slowly becoming more reliable and repeatable than an LLM. AI has been critical to managing and testing these huge rulesets efficiently, but surpisingly minimal value at runtime, and its very slow.
• User-facing configuration management: Our configuration sets are very, very complex, and there’s a very steep learning curve to modifying them. We tried to put AI agents in front of it for users, but fundamentally, it was still slow, still hard to grasp, and still unworkable to get most users comfortable with it. So instead we built simpler, more intuitive UI layers over a subset of the configuration, and have users work with real humans when they need more complex workflows.
• Quality checks: We’ve started replacing more and more of the quality process with deterministic, configurable rules — these are more interpretable, more reportable, and quickly becoming more accurate than an LLM, which truly doesn’t understand the domain nearly as well as people on the ground. Not to mention being much cheaper.
• User training: Traditional user training tools still work — help tooltips, feature search, interactive tutorials, cheat sheets, videos, etc., are far faster, friendlier, and more efficient to use than a chat window 95% of the time. AI’s job is to help build the trainings in advance, not try to build them live, custom for each user.

Builder workflows where AI becomes more valuable the more we use it

• Core software development: Writing code with AI in 2026 is a completely different experience than writing code by hand. And when architected with code replaceability in mind, rather than reusability, it is a much better experience, especially for the replaceable parts of the system: UI widgets, simple backend API routes, hardware integrations, etc.
• Admin assistants: Agents given to admins to manage complex configuration — unlike operational users, who need to insert their new configuration in seconds or they give up, business administrators have the time to make something good, and to talk with an AI agent, understand the complexity, and make it good. Their day-to-day workflow is much more like a developer’s, where they are trying to build plans and projects.