Writing
Technical writing on computer use agents, self-healing RPA, and healthcare EHR automation. From the team building Laminar.
Architecture
Why We Dropped the Planner from Our Agent Architecture
Most computer use agents use a planner-executor hierarchy. Modern LLMs are good enough that the extra layers are now overhead. Here is what we learned when we simplified.
Split Reasoning from Grounding: Why Two Models Beat One
Asking one model to both decide what to click and predict where it is on screen is asking it to do two very different jobs at once. We found that splitting them changes everything.
Why Every Action Needs a Verification Step
One missed click without a check cascades into five more bad actions. One missed click with a check gets caught in three seconds. The math is simple.
Your Agent Should Get Faster the More It Runs
The first time an agent navigates a system, it explores. The hundredth time, it should know the path. Most agents do not work this way.
When Clicking Is the Wrong Answer
Some tasks are fundamentally computational. Summing a spreadsheet column, parsing a PDF, renaming files. Clicking through a GUI for these is slow and fragile. The agent should know when to write code instead.
Infrastructure
Running RPA at Scale Is a Distributed Systems Problem
You have a pool of Windows VMs. Requests come in through an API. Something needs to route, queue, health-check, and retry. This is not an automation problem. It is an infrastructure problem.
The Hidden Complexity of Windows VM Session Management
Session timeouts, memory leaks, OS updates at 3am, MFA prompts. Your automation can be perfect. The Windows environment will still break it.
Routing Desktop Workflows Across a Pool of VMs
One API call triggers execution. The platform handles routing, session state, and failover. Here is what that orchestration layer actually needs to do.
Healthcare
What Makes Healthcare EHR Automation Different
The stakes are different. The UIs are deceptive. Session management is brutal. Here is what you learn when you automate EHRs in production.
Why AI Companies Lose Deals (It Is Not the AI)
The AI works. The model generates the note, processes the invoice, classifies the document. Then the output needs to go into a legacy system with no API. That is where deals die.
Automating Clinical Documentation Without Breaking the Chart
Writing notes into an EHR is not a text-entry problem. It is a navigation and verification problem with text entry in the middle.
The 30-Second Constraint in Clinical Workflows
A doctor finishes a visit. The note needs to be in the chart before the next patient walks in. Most automation architectures cannot hit this window.
Industry
The 10/90 Rule of RPA: Building Is Easy, Maintaining Is Everything
Building an automation takes a week. Keeping it alive over 12 months is where all the time goes. The ratio is roughly 10% build, 90% maintain.
The Compound Failure Math That Makes Click Accuracy Critical
95% per-click accuracy sounds great. On a 20-step workflow, it means two out of three runs will have at least one misclick. Here is why recovery matters more than accuracy alone.
The Three Hardest Things About Production RPA
They have nothing to do with the automation itself. Detection, investigation, and ongoing resilience are what consume your engineering hours.
Silent Failures: When the Automation Runs Fine But the Data Is Wrong
The worst kind of failure is not a crash. It is when the automation completes 'successfully' and puts data in the wrong field. Nobody finds out for hours.
Why Hard-Coded RPA Breaks Every Time the UI Changes
Traditional RPA memorizes element positions and IDs. Computer use agents look at the screen. The difference matters every time a vendor pushes an update.
Platform
Going from Zero to Production in Three Weeks
Our customers go live in three weeks, not three months. Here is what that timeline actually looks like and what makes it possible.
Observability for Desktop Automation: Why Logs Are Not Enough
When a desktop automation fails, you need to see what the screen looked like, not read a log file. Visual replay changes how fast you can investigate and resolve issues.
The Long Game: From Integration Layer to Something Bigger
Every workflow automated captures data about how legacy systems actually work. Over time, that understanding compounds into something more valuable than the integrations themselves.