Paul Karsten 11/21/25 Paul Karsten 11/21/25

Shadow AI

Employee use of unauthorized AI tools is inevitable. Rather than blocking it, organizations should investigate why it's happening and build governed internal AI platforms that meet those needs. Success requires permission-aware data access, automated security controls, and treating AI governance as product development, not just compliance. Transform the risk into a competitive advantage.

Paul Karsten 11/4/25 Paul Karsten 11/4/25

AI - The Rote Machine

AI's perceived intelligence is often overstated. It is only as “intelligent” as the data it's trained on. Poor quality and unfit data lead to “incorrect”, unsupported, and potentially reputational affecting decisions. Actual progress in AI requires a relentless focus on data fitness, quality, governance, and fairness.

Paul Karsten 11/4/25 Paul Karsten 11/4/25

SRE Wasn't Invited to the AI Party

There is a significant disconnect between the push for AI adoption in leadership and its practical application within Site Reliability Engineering (SRE) and infrastructure teams. While developers benefit from AI tools like Copilot for code completion and testing, SRE teams, whose work involves declaring desired states, orchestrating systems, and troubleshooting unique infrastructure challenges, find current AI tools largely unhelpful.

AI could make a difference in SRE by acting as intelligent agents that correlate logs, analyze metrics, and identify patterns during incident response, thereby reducing Mean Time To Resolution (MTTR) and demonstrating tangible business value, rather than focusing on traditional code-centric productivity metrics.

Paul Karsten 10/24/25 Paul Karsten 10/24/25

The Velocity Trap

The AI revolution is sparking a familiar refrain: "How do I know that developers are being more productive?" Because Velocity is easy to measure, it is the metric I hear discussed most often. Velocity may be easier, but value, while more difficult, is the only thing that matters.

Paul Karsten 10/22/25 Paul Karsten 10/22/25

5 Key Indicators That AI Isn't The Answer

This post outlines five key considerations for deciding whether to pursue AI: the existence of clear rules, data volume and variability, significant regulatory requirements, static environments, and unclear ROI. AI should be a multiplier for disciplined thinking and well-designed processes, not a replacement. Focusing on data quality and automation maturity should precede any AI adoption.

Paul Karsten 10/21/25 Paul Karsten 10/21/25

Compliance, Coffee, and Machine Learning

Large Language Models (LLMs) can transform compliance document review from a slow, error-prone and manual process into a scalable, efficient operation. Unlike traditional automation that relied on rigid keyword matching, LLMs understand context and nuance, and when trained on organization-specific examples—such as approved contracts and compliance templates—they can assess documents against internal governance standards.

Paul Karsten 9/11/25 Paul Karsten 9/11/25

Is AI the new Lotus Notes?

AI driven development is creating a "Citizen Developer" movement that, while promising innovation and agility, risks repeating the "technical debt" problems seen with Lotus Notes in the 1990s. This new era of development can lead to undocumented, and unmaintainable and significant security and compliance vulnerabilities, as business units bypass central IT for quick solutions. To mitigate these risks and foster this innovation, organizations should implement proactive strategies like robust developer platforms with effective guardrails to bridge the gap between IT expertise and business user needs.

Greg Pearman 9/10/25 Greg Pearman 9/10/25

Citizen Coder in your Enterprise

At Parallaxis, we recognize "Citizen Coders" and "Citizen Developers" as the driving force behind what the broader industry calls "vibe coding"—the intuitive, rapid development approach that prioritizes getting things done over traditional methodologies.

Paul Karsten 9/9/25 Paul Karsten 9/9/25

A Coalition of the Motivated

A Community of Collaboration approach fosters genuine problem-solving, increases drive and momentum, and ultimately leads to better business results and increased retention. By creating cross-disciplinary communities of affected individuals focused on specific problem statements with defined outcomes we drive greater innovation. They should be temporary in nature and empowered to rethink processes.

Paul Karsten 4/17/25 Paul Karsten 4/17/25

The Illusion of Just Knowing

Introducing AI/ML in your business emphasizes the importance of data for understanding business operations and driving growth. Relying on assumptions hinders progress and creates an illusion of competence. The document advocates metrics, experimentation, and the scientific method to create a data-driven approach to doing business.

Paul Karsten 4/9/25 Paul Karsten 4/9/25

There is no Bad Data

Data's value depends on its intended use. Operational data collection often prioritizes transactions over analysis, resulting in data not optimized for later purposes. Technical data aggregation can introduce biases. Unclear business requests and data silos complicate analysis. To leverage data effectively, we need to be flexible on how we analyze the data we have at hand.

Paul Karsten 2/17/25 Paul Karsten 2/17/25

Ethics Is Not “One and Done”

Implementing ethics in Machine Learning is not a one-and-done effort. Organizations must build trustworthy systems that serve their customer’s best interests fairly. Ultimately, ethically designed machine learning models align with compliance and regulations while also avoiding harmful outcomes.

Paul Karsten 10/28/24 Paul Karsten 10/28/24

Practical Business Reasons to Resist the Allure of AI

There are many traps along the journey required to leverage AI/ML to generate value for your business. Success relies on aligning AI/ML initiatives with clear business objectives and understanding their true potential.

Paul Karsten 10/7/24 Paul Karsten 10/7/24

Anti-Patterns in Data Mesh

This article explores common anti-patterns in implementing Data Mesh, a decentralized data architecture emphasizing domain-oriented data ownership. While Data Mesh aims to enhance data accessibility and usability across organizations, its success relies on understanding core principles: domain-driven data ownership, data products, and federated governance.

Paul Karsten 9/23/24 Paul Karsten 9/23/24

Data Mess to Data Mesh

The standard strategy of centralizing data into a single repository often leads to chaotic "data swamps.” Due to poor data quality and governance issues, these swaps hinder efficient analysis and decision-making. An alternative approach, known as Data Mesh, proposes a decentralized architecture focused on treating data as a product.

Paul Karsten 9/9/24 Paul Karsten 9/9/24

Transformative Data Pipelines for Analytics Using AWS Glue

Practical considerations for building analytics-ready data pipelines and data products using AWS Glue with Jupyter notebooks, Python, and Terraform.

Paul Karsten 8/30/24 Paul Karsten 8/30/24

MLOps Automation

MLOps requires specialized knowledge that traditional DevOps teams lack. The challenges related to data quality, consistency, and accessibility demand a different set of skills and tools.

Paul Karsten 8/23/24 Paul Karsten 8/23/24

Model Release & Assessment Phase

This 3rd phase of our Data Science Process explores the release of ML models into production and the importance of ongoing monitoring and assessment.

Additionally, it provides a framework for defining "done" and achieving a high-quality model release.

Paul Karsten 7/10/24 Paul Karsten 7/10/24

Model Development

This blog post outlines the second phase of our Data Science Process: Model Development. Which involves building, training, and evaluating models based on data gathered during Question Formation. The process is iterative, experimenting with different algorithms, features, and parameters in a sandbox environment before scaling to larger datasets. Model performance is evaluated using metrics, validation for overfitting/underfitting, and checks for robustness and interpretability. Finally, models must be versioned, monitored for data drift, and continuously updated to ensure they remain effective and relevant over time.

Paul Karsten 5/13/24 Paul Karsten 5/13/24

Question Formation and Data Analysis in Data Science

This blog post focuses on the first phase of our Data Science Process: Question Formation and Data Analysis. In this phase, we iterate multiple times through question formation, data collection, and exploration. Initial questions are likely to be of low fidelity. Through the process of data exploration, the questions gain fidelity and drive toward business value.

Welcome to the Parallaxis Blog