Daily Tech Digest by Kannan Subbiah: private AI

Quote for the day:

"When you want to succeed as bad as you want to breathe, then you’ll be successful." -- Eric Thomas

🎧 Listen to this digest on YouTube Music

▶ Play Audio Digest

Duration: 18 mins • Perfect for listening on the go.

‘Botsitting’: The AI time-savings killer only governance can stop

While artificial intelligence promises to free up employees for valuable tasks, a recent study reveals that workers lose more than half their saved time to “botsitting.” Digital workers save roughly eleven hours a week using these tools, but spend over six hours managing them—providing missing context, checking outputs, fixing mistakes, rewriting prompts, and correcting inaccurate answers. As a result, businesses are missing out on the full return on their investments. A core issue is poor governance and a lack of training. Employees often use AI for simple tasks like drafting emails, distrusting it for complex work. Moreover, there is “coordination neglect,” where an individual’s productivity gains create unexpected work for others downstream. For instance, when workers pass along unchecked, AI-generated content, teammates must spend unbudgeted time cleaning up the mess. Experts warn that simply implementing tools without clear guidelines on verification processes and data context leads to inefficiency. To truly benefit from these technologies, organizations must focus on proper deployment, establish clear oversight, and define quality standards rather than merely counting how often tools are used. Reliable outcomes require thoughtful management, not just fast adoption.

The database that refused to die: How Postgres survived its own creators

Postgres, one of the world's most widely used database systems, began its life with an uncertain future. Created by database pioneer Michael Stonebraker in the 1980s as a successor to Ingres, the project was essentially abandoned by its creator in the mid-1990s. Instead of fading into obscurity, Postgres was rescued by a dedicated community of independent open-source volunteers. These contributors preserved Stonebraker's foundational, highly adaptable architecture—which allowed for complex, user-defined data types rather than just basic strings and numbers—while adding standard SQL capabilities. Today, this collaborative rescue effort has established Postgres as a cornerstone of modern cloud computing infrastructure. Its enduring success stems from its foundational design philosophy. While proprietary database systems traditionally optimize their software to suit the specific needs of massive enterprise clients, Postgres was built to handle the diverse workloads of general users. By seamlessly accommodating complex data formats like geographic information and computer-aided design files, it solved real-world problems for a broad audience. Ultimately, the survival and widespread adoption of Postgres demonstrate the power of open-source software, proving that community-driven development can outlast even the original creators to become a resilient industry standard.

Why private AI is the smarter bet

Although many businesses initially assumed artificial intelligence would naturally live in the public cloud, reality is forcing a shift toward private, on-premises systems. According to the article, this transition stems from growing concerns about uncontrolled costs, security vulnerabilities, and operational fit. As companies move from small experiments to organization-wide implementation, the pay-per-token pricing models of public cloud providers risk becoming massive utility bills that wipe out business gains. Consequently, the future of enterprise AI leans toward a hybrid model. Rather than relying entirely on giant public models, businesses are discovering that smaller, specialized AI models can handle tasks better while running closely to their own private data. This approach offers better control over predictable workloads and eliminates surprise expenses. Furthermore, keeping AI in-house strengthens security and data governance. Using public AI tools raises the real danger of employees inadvertently exposing sensitive or proprietary information. While building and managing private AI networks requires significant investment, skill, and discipline, the long-term benefits of controlled costs, tight security, and owned infrastructure make it a much smarter choice for major production workloads.

AI Cost, Security Pressures Push Enterprises Toward Private Cloud, Broadcom Says

According to a recent report from Broadcom, organizations are increasingly moving their artificial intelligence operations away from public cloud services and toward private cloud setups. As businesses shift from merely testing artificial intelligence to running real-world applications, they are discovering that private networks offer better handling of costs, security, and data control. The study reveals that over half of surveyed enterprises now plan to run their active intelligence systems on private infrastructure. Meanwhile, public cloud usage for these specific tasks has dropped notably over the past year. Interestingly, cost management has now surpassed security as the primary concern with public platforms, as business leaders face unpredictable pricing for computing power and data storage. Because of this, more than eighty percent of companies are either moving or considering moving their systems back in-house. While public networks remain useful for basic testing and flexible storage, the heavy demands of daily production require a more stable environment. Strict data privacy rules further encourage this transition. Ultimately, businesses are finding that dedicated internal systems provide the financial predictability and reliable protection necessary to safely grow their technological capabilities.

How to Modernize Legacy Applications Without Disrupting Business

Upgrading older software systems is a pressing challenge for modern organizations. Delaying these updates can hinder new capabilities, consume vital budgets with maintenance costs, and create risks as experienced programmers retire. However, many companies hesitate because poorly planned upgrades often cause severe business interruptions. To avoid taking systems offline, experts recommend a gradual approach rather than attempting a risky, sudden replacement. This method relies on careful planning and proven structural designs. For example, organizations can build new services around the existing system, slowly routing traffic to the new components as they are tested and proven. Another reliable method involves running both the old and new systems at the same time to ensure they produce identical results before fully switching over. It is also important to use a translation layer to prevent the flaws of the old data formats from infecting the new setup. A successful upgrade generally follows a structured path: assessing current dependencies, planning the target design, running a small initial pilot, scaling the effort across other applications, and maintaining ongoing oversight. By strictly adhering to these methods, businesses can confidently update their technology and maintain continuous daily operations.

Data Lakehouse Architecture Layers: AI Needs More Than Just Infrastructure

Organizations have invested heavily in data lakehouses to store and process large amounts of information for analytics and artificial intelligence. While these setups handle storage and compute well, they often fall short in practical application. Data remains scattered across different cloud environments and operational systems, meaning business teams and AI models still struggle to access reliable information without technical assistance. The fundamental issue is no longer about where data is kept, but how it is connected and understood. AI tools, in particular, require more than just raw data; they need clear context and strict governance to function accurately and safely. To solve this, a new logical layer is emerging in data architecture. Instead of replacing the lakehouse, this access layer sits on top of it. It connects distributed information, applies consistent rules, and provides clear meaning to the data without requiring it to be moved or duplicated. By pairing traditional storage with this new governance layer, businesses create a stronger foundation. This approach reduces friction, ensures that both human users and systems have the context they need, and allows organizations to focus on practical outcomes rather than managing complex infrastructure.

The Four Elevations of Effective Fraud Prevention

Effective fraud prevention requires more than just checking individual steps; it demands a layered approach to monitor customer behavior comprehensively. To build a resilient defense, organizations should evaluate activities across four key elevations. First is the transaction level, which looks at single interactions like logins or purchases. While important, relying on this alone can miss larger patterns because attackers frequently change their tactics. The second elevation is the account level, where monitoring a user's behavior over time helps distinguish normal activity from suspicious anomalies, such as sudden changes to contact information or unusual transfer requests. The third elevation expands to the platform level, allowing teams to analyze trends across all grouped accounts. This broad view helps quickly spot coordinated attacks or fraud rings sharing the same devices or geographic locations. Finally, the network level involves collaborating with external data providers to share insights across different companies, ensuring that a threat detected by one organization is immediately known to others. By integrating these four perspectives, businesses can confidently identify complex fraud schemes early, reduce false alarms for legitimate users, and secure their operations without disrupting the everyday customer experience.

Bridging the gap between leadership's AI enthusiasm and employee pushback

Corporate leaders and everyday employees often view artificial intelligence through entirely different lenses. While executives and board members see AI as a path to efficiency, cost reduction, and innovation, employees frequently view the technology with caution. Many workers worry that AI will result in job losses, create mentally exhausting workloads, enable invasive workplace surveillance, and harm the environment. Chief Information Officers (CIOs) find themselves caught in the middle and must bridge this divide. If IT leaders ignore workforce anxieties and force AI integration, they risk damaging company morale, losing valuable talent, and wasting money on tools that employees simply refuse to use. To resolve this tension, CIOs need to look beyond basic financial metrics and instead measure actual employee sentiment and tool usage. Having open, honest conversations with staff about their fears is essential. By creating a culture where workers feel safe sharing their concerns, companies can build trust and ease anxiety. Rather than rolling out technology blindly, leaders should clearly communicate the company's AI strategy and empower early adopters to guide their peers, ensuring the transition supports both business goals and the well-being of the team.

AI Works, Pull Requests Don’t: How AI Is Breaking the SDLC and What To Do About It

In the presentation "AI Works, Pull Requests Don't," Michael Webster examines how the rise of artificial intelligence coding assistants is severely straining traditional software development lifecycles. While AI tools initially act as powerful amplifiers that can increase development speed by three to five times, this burst in productivity is often temporary. Developers and AI agents are generating massive amounts of code, sometimes adding twenty-five times more code than they delete. As a result, human reviewers are overwhelmed by enormous pull requests, creating significant bottlenecks in the review process and leading to a steady accumulation of technical debt. Drawing on queuing theory, Webster explains that delays inevitably occur when the rate of incoming code surpasses the team's capacity to process and review it. To resolve these challenges, engineering teams must adapt their validation pipelines. He recommends implementing test impact analysis, a method that runs only the tests affected by recent code changes rather than the entire test suite. By relying on automated validation tools to quickly verify AI-generated output, teams can successfully maintain software stability, reduce testing costs, and manage the high volume of code without sacrificing overall quality.

Hackers Exploit Weak Credentials and Internet-Facing PLCs to Breach Water Utilities

Water and wastewater utilities across the United States and Europe are facing increasing threats from state-sponsored groups affiliated with Iran, Russia, and China. Rather than relying on complex software, these attackers exploit fundamental security oversights, like internet-exposed control systems, default passwords, and inadequate network separation. This shift indicates that targeting civilian infrastructure has become a deliberate method to test emergency responses, create public anxiety, and position adversaries for future conflicts. For instance, Iranian-linked groups have used factory credentials to access unprotected systems, while Russian-affiliated actors actively disrupted operations by overflowing water tanks in Texas and opening floodgates in Norway. Meanwhile, Chinese groups take a quieter approach, establishing long-term access within utility networks to maintain leverage for potential disputes. To counter these vulnerabilities, security experts advise facility operators to implement basic defenses immediately. These include removing physical control systems from direct internet exposure, enforcing strict login requirements, replacing default passwords, and firmly separating industrial equipment from standard computer networks. By addressing these entry points, utilities can effectively reduce their risk of compromise and safely protect vital public water resources from further interference.