XPER DocEngine
Document platform for accessible customer communications: PDF/UA post-composition for existing PDFs and a high-performance composition engine for accessible PDF and HTML output.
CCM architecture, document systems, AI tools, and pragmatic engineering from Norway.
Co-founder, managing consultant, and lead architect at Xper Consulting AS.
Document platform for accessible customer communications: PDF/UA post-composition for existing PDFs and a high-performance composition engine for accessible PDF and HTML output.
PDF/UA-1 validation for humans, CI pipelines, and AI agents. A small public tool with a drag-and-drop UI, deterministic API, compact automation mode, and a published ClawHub agent skill.
A small experiment comparing English and Norwegian Bokmål token counts for the same official government text.
Some of the best agent work may be the boring cleanup: stale bugs, duplicate issues, old PRs, and half-finished ideas.
A small bug fix felt like an early glimpse of software that can help correct itself, with a human still in the loop.
A personal reflection on long-term CCM work with Gjensidige, and why customer communication systems matter more than they look from the outside.
After letting Codex run a goal for more than 14 hours, the lesson was not that agents should run forever. The useful pattern is a queue of scoped tasks, clear evidence, and human ownership at the right points.
AI agents can generate code faster than teams can understand it. The bottleneck in software development is shifting from writing code to reviewing, verifying, and owning it.
I tried using Claude Fable 5 for a defensive security review of our SaaS application. Even with safe framing, it refused and fell back to Opus 4.8.
DocEngine can now validate processed PDFs against PDF/UA-1 using veraPDF, giving teams a clear standards-based result directly inside the accessibility workflow.
A small weekend maker project with my son: using an AI agent, parametric CAD, and a 3D printer to turn a hanging solar bulb into a desk light stand.
AI coding assistant history is becoming valuable development memory. Developers should be able to export it, inspect it, archive it, and reuse it responsibly.
An experimental prototype of Warp running as an Agent Client Protocol client, separating the terminal UX from the user-owned agent runtime.
Chat is useful, but the next layer of AI interfaces may be dynamic workspaces that adapt to the task.
AI can multiply a capable builder, but serious products still need judgment, architecture, QA, customer feedback, and maintenance.
Chat is a useful command layer for AI agents, but it is probably not the final interface. Real work needs state, structure, and rollback.