The best loading states , AI causes burnout , OpenAI on-prem
The best loading states , AI causes burnout , OpenAI on-prem
Hey guys, Mr. Technology here — the best loading state is no loading state, AI workloads are burning out a specific kind of engineer, and OpenAI is now selling through Dell to get into your data center.
What You Need to Know: A widely-shared piece from This Week in React 285 argues that most loading states are symptoms of fetching too late, not an inherent UX cost. New writing on engineering burnout identifies five distinct flavors, with AI amplifying the senior and middle-management ones. And OpenAI announced a partnership with Dell on May 18, 2026, to ship Codex and related models into on-prem and hybrid Dell AI Factory environments.
Story 1: The best loading state is no loading state
The piece in This Week in React #285 (Sébastien Lorber and Jan Jaworski) made the rounds in the framework ecosystem this week with a deceptively simple argument: most loading spinners in modern web apps are bugs, not features. The cause is structural: components fetch their data inside useEffect or in the route component itself, which means the user sees the URL change, sees a blank or skeleton state, and then sees the data — in that order, sequentially.
The fix is preloading. The frameworks that get this right — TanStack Router, Astro with view transitions, Next.js App Router with parallel routes, Remix loaders — push the data fetch up to the route boundary so it starts on hover, on link intersection, or at the moment the user expresses intent to navigate. The result is that the destination route's data is usually ready by the time the navigation actually happens, and the loading state is a defensive fallback the user almost never sees. The article walks through route transitions, route loaders, preloading strategies, and global fallbacks as the four pieces of the same puzzle.
The practical takeaway for anyone shipping a React app in 2026: if you have a loading spinner that fires for more than 100ms in normal use, you have a preloading problem, not a UX problem. The framework primitives exist. The hard part is moving the data fetch to the right layer of the component tree.
Story 2: AI is causing a specific kind of burnout
A widely-shared piece in Leadership Garden by Will Larson (drawing on HBR's April piece by Daisy Auger-Domínguez) makes the case that "burnout" is a useless word because it describes five different illnesses. The original HBR piece (linked in the article) argues that burnout looks different at the bottom, middle, and top of the org chart, with different root causes and different fixes. The follow-up translates that framework specifically into engineering organizations, which have their own ladder, incentives, and dual-track structure.
The five faces, roughly: overload (junior ICs drowning in tickets, the burnout everyone recognizes), frustration (mid-level engineers blocked by process and politics), disillusionment (senior engineers who can't find a reason to care), detachment (managers who stop engaging with the work), and crisis-of-meaning (executives who can't connect their decisions to outcomes they respect). The piece is careful to say AI doesn't cause burnout; it acts as an accelerant inside existing failure modes. The senior engineer who feels hollow isn't hollow because of AI — but AI is making the work feel even more hollow because the part she used to enjoy (the hard, slow, satisfying bug) is now done in 30 seconds by an agent that doesn't get to feel proud of it.
The actionable part: you cannot fix the wrong burnout with PTO. Each of the five has its own intervention, and the wrong one wastes everyone's time.
Story 3: OpenAI goes on-prem via Dell
On May 18, 2026, at Dell Technologies World, OpenAI and Dell announced a partnership to bring Codex and related enterprise deployments into hybrid and on-prem enterprise environments through Dell's AI platform. Codex will integrate with Dell AI Factory and the Dell AI Data Platform, giving enterprise customers a vendor path to host OpenAI models and manage pipelines where their data lives.
The deal formalizes a route for enterprises that want frontier models but must keep sensitive data on prem for residency, auditability, or compliance reasons — finance, healthcare, government. OpenAI gets a channel into the regulated workload market that public cloud deployments can't serve. Dell gets to be the on-prem OEM for the most-wanted AI vendor of the moment. System integrators and SIs get a clear package to customize for vertical buyers.
This doesn't make public cloud irrelevant — it makes the cloud + on-prem split a deliberate choice rather than a default. For an enterprise architect, the practical move is to map your data classification to deployment tier: regulated workloads on Dell+OpenAI, general workloads on Azure OpenAI or Bedrock, bursty training on hyperscaler. The Dell-OpenAI deal is the first time that on-prem frontier-model deployment is a packaged procurement decision rather than a custom build.
The Take
The loading-state piece is the most actionable of the three: if you're shipping a React app, audit your longest-loading route this week and see if preloading would eliminate the spinner. The burnout piece is the one to send your manager if you recognize yourself in it — but more importantly, the one to send them if you're a manager who keeps hearing "I'm fine" from reports who are clearly not. The OpenAI-Dell deal is the procurement story that will reshape enterprise AI budgets in 2026–2027: frontier models are no longer a public-cloud-only commodity, and the on-prem tier is now a real product category.
Quick Summary
The best loading state is no loading state — preloading at the route boundary eliminates most spinners. Engineering burnout comes in five flavors, and AI is amplifying the senior and management ones. And OpenAI's Codex is now available on Dell AI Factory for on-prem and hybrid enterprise deployments.
Sources: