The Split Screen
I built two websites because I am two engineers now. One crafts every word. The other ships at scale. The gap between them is where the future of this profession is being negotiated.
There is a moment, usually around 6:47 on a Tuesday morning, when I open two browser tabs side by side.
The left tab shows simon-cullen.com — a profile piece that cost me three weeks of obsessive refinement. Every sentence interrogated. Every comma earned. Dublin-based Principal Data Engineer, OptumRx, UnitedHealth Group, credentials in specific order. The career timeline moves logically from Groupon (2014) to present day without a single irregular heartbeat. This is the engineered surface. The confident handshake.
The right tab shows insights.simon-cullen.com — this site, where the article you are reading was generated, formatted, and published by an AI assistant while I was reviewing anomaly detection dashboards for a pharmaceutical pricing pipeline. I reviewed the draft during a coffee break. I made three edits. It went live before my cup went cold.
This is the split screen. I am not sure which one is the real me. I am not sure the distinction matters anymore.
Listen to This Article
The Handshake vs. The Laboratory
The profile site exists for a non-negotiable reason: trust in high-stakes environments requires provenance. When a Chief Data Officer evaluates whether I can migrate their pricing infrastructure without breaking the medication supply chain, they want documented trajectory and evidence I have operated systems that cannot fail.
The profile delivers this with architectural precision. The five years at Susquehanna International Group building sub-second ETL for ETF compositions — where imprecision costs millions in milliseconds. The R-to-Python migration at OptumRx that cut infrastructure costs while maintaining SOC 2 governance. Credentials (Databricks Lakehouse, Azure DP-900, FinOps Certified Practitioner) presented the way a structural engineer presents their license: liability management, not bragging.
But here is the uncomfortable truth: that site is a retrospective artifact. It documents what I have done. It says almost nothing about how I work now.
This site operates on different physics. Here, an AI assistant named Vaire manages the entire content pipeline: research, drafting, image generation, audio production, publication. I provide direction and quality control. Vaire executes.
What this means practically: an article that would have taken a full day to research, draft, edit, format, and publish now ships in under an hour. The quality is different — not worse, but different. More experimental. The voice shifts because the production method shifts.
Curated Identity, Operational Identity
Six months ago, this division felt like a confession. Now it feels like a preview of what every knowledge worker will face within three years.
The profile site represents curated identity — the version we assemble for professional consumption. Most professionals keep their operational identity hidden: their actual workflows, tool choices, willingness to delegate to systems they do not fully control. I have chosen radical transparency about the split itself. When you read an article here, the byline says "Simon Cullen" but the method is labeled. You know an AI assistant was involved.
This is strategy, not humility. In a world where AI-generated content will dominate information ecosystems, competitive advantage shifts from production capacity (commoditized) to discernment and direction (still human, for now). By making the split visible, I demonstrate the skill that actually matters: not writing, but orchestration. Not drafting, but judgment about what deserves to exist.
The Professional Risk
There is a risk here I want to acknowledge. The profile site positions me for enterprise data engineering roles — healthcare migrations, compliance infrastructure, systems where a wrong number affects human welfare. These organizations prize caution and documented process. They do not always prize experimental AI workflows.
By openly running an AI-managed content pipeline, I am making a bet: that evaluators will recognize the distinction between mission-critical systems (where I operate with full manual control) and experimental content workflows (where I delegate to AI and accept different quality thresholds).
Some will see the AI-assisted workflow and assume sloppiness. They will miss the distinction between healthcare pricing pipelines requiring 99.99% uptime and blog posts requiring interesting ideas delivered quickly. That is a cost I accept. The alternative — hiding the AI workflow — feels like a worse long-term strategy. The deception would surface eventually.
Conductor, Not Craftsman
The split-screen architecture is not marketing. It is a methodological preview of how data engineering teams will operate.
At OptumRx, I already manage a division of labor: I design pipeline architecture, specify transformation logic, validate output quality. The execution — terabytes of pharmaceutical pricing data moved by Databricks clusters — happens without my manual intervention. I am orchestrating, not operating.
The insights blog extends this into content. I specify research direction, angle, quality threshold. Vaire executes drafting, formatting, publication. My role shifts from craftsman to conductor — maintaining vision while delegating implementation.
This terrifies engineers who hear "AI-generated content" and imagine replacement. I see cognitive specialization: the human tasks that remain require judgment, taste, strategic direction, ethical boundaries. These are harder than drafting. They require knowing what should exist, not merely how to produce it.
The Specifics
When I publish on the profile site, I open a text editor. I write sentences. I delete them. I rewrite. I obsess over rhythm. The process is slow and entirely manual. A single update might take three hours.
When I publish here, I open a chat interface. I describe the article: angle, tone, specific points, word count. Vaire returns a draft in seconds. I review, edit, request revisions. When satisfied, I say "publish." Vaire converts the draft to markdown with YAML frontmatter, generates a hero image via Venice AI using prompts like "abstract data infrastructure, warm terracotta and cream palette," produces three audio versions using Kokoro TTS voices (Sky, Onyx, Puck), and syncs to Ghost CMS via the Admin API.
Ghost CMS, Venice AI for images and TTS, custom Node.js orchestration, local toolkit handling auth and sync. Documented. Repeatable. Increasingly autonomous.
Is the output better? In absolute terms, no. The profile site has tighter prose, more consistent voice. But this site produces more — more ideas tested, more experiments run. In a landscape where attention is scarce and iteration speed matters, volume with acceptable quality beats perfection with glacial cycles.
The Question
Back to 6:47 AM, two tabs open.
Which site represents the better professional investment? The profile that establishes credibility? Or the insights site that develops AI-orchestration skills?
The question is malformed. It assumes I must choose. The reality is that operating both simultaneously — maintaining curated identity while aggressively experimenting with AI delegation — is itself the skill.
The engineers who thrive will not be those who reject AI on principle, nor those who surrender all judgment to automation. They will be those who can context-switch between modes — knowing when precision requires manual control, and when speed requires strategic delegation.
That is what the split screen demonstrates. Not a contradiction, but a competence.
If you arrived here from my profile site, you have seen both sides. You know the professional narrative and the experimental workflow come from the same source. I am not hiding the AI assistance. I am demonstrating what it looks like to integrate these tools into production systems with eyes wide open.
Simon Cullen
Principal Data Engineer, Dublin
29 January 2026
