To charge on with our new Workflow Wizardry, let’s get crackin’ on The Cost-Cutting AI Super-Stack!

The Comet’s Tale ☄️ — Issue #011

Your Daily Beacon for All Things Comet, Perplexity, and AI Advantage

Greetings, Astral Adventurers! Chris here. 👋🏼

Continuing our celebrated Workflow Wizardry Week, we’re teaching you how to build automated workflows that turn manual tasks into background brilliance, today is about the inevitable next step: workflows that learn, evolve, and repair themselves. We are officially moving from automation to autonomy.

For years, coders and builders have dreamed of a codebase that gets smarter the more it runs. Thanks to Perplexity Labs and its advanced reasoning models like Sonar Reasoning Pro, that sci-fi vision is rapidly becoming a reality. We’re talking about tools that don't just execute code but also audit it, debug it, and then implement the enhanced version—all while citing the original sources that informed the repair.

Buckle in. Today, we’re unpacking my agentic coding experiment, codex-beta, landing me 4,500 impressions and growing 😱. Folks, I was not anticipating another Gemini-flow like interest, but I appreciate and 💙 y’all for it!!! So in that spirit, we’re derailing #WizardingWorkflowWeek once again, and introducing the concept of self-evolving software, and showing you how Perplexity tools are the launchpad for this new era of engineering.

I didn’t know THIS would STILL BE GOING… 🤯🤯🤯

In case you missed it: What actually IS Perplexity Comet? ☄️

Simply put, Comet is Perplexity’s AI-powered browser with your built-in research assistant—The Comet Assistant. It’s a unified AI search engine with instant context and automation across every site.

Comet accomplishes the following:

• Instantly reads and digests web pages for you.

• Cross-examines facts from multiple sources, delivering real-time, trustworthy fact-checks.

• Helps you shop smarter by checking price history and analyzing reviews.

• Drafts emails, documents, and summaries with context.

• Answers questions about anything you see—right on the page.

The Comet Advantage: While ChatGPT draws from past data, Comet and The Comet Assistant operate in the now—using live web searches to ground answers and insights.

🚀 Perplexity/Comet News of the Moment

Featured Update: Labs Unlocks Agentic Code Self-Correction

Perplexity Labs, the dedicated workspace for turning ideas into reality (reports, dashboards, simple web apps), has just rolled out a major feature allowing advanced integration with user-hosted code repositories. This dramatically accelerates the path toward truly autonomous software development.

Why this really matters:

1. Code Execution and Audit: Labs can now not only write and execute code (which it already did for data structuring and charting) but also monitor its performance during runtime and identify failure points.

2. Self-Correction Loop: Using its Deep Research and Reasoning models (like Sonar Reasoning Pro), the platform can search for best practices or bug fixes, generate a patch, and test the augmented code agentically—providing a report on the efficiency gain.

3. The New Engineering Paradigm: This positions Perplexity Labs not just as a tool for creating code assets, but for managing the life cycle of simple applications, delivering what looks and feels like a dedicated AI team.

🌠 Project of the Day: Self-Evolving Code in the Wild

In this spirit of autonomous progress, I’m sharing the latest update from my experimental GitHub repository: https://github.com/clduab11/codex-beta.

The Mission: To take this bare repository that started out as just a way of taking https://github.com/clduab11/gemini-flow and https://github.com/ruvnet/claude-flow to build upon OpenAI’s codex; and then 1) gpt-5-codex with reasoning dropped, and 2) it’s able to automatically monitor its own core functions (in this case, self-developing for stellar demand for codex-beta [just a temporary name, I promise!], Parallax Analytics) and self-deploys code fixes if Python/TS linting fails below a certain threshold.

The Analogy: If conventional coding is like building a car and handing over the maintenance manual, Agentic Self-Evolvement is like building a self-driving car that carries its own tool kit and constantly rewrites its own maintenance manual while driving.

The Breakthrough: Using a persistent SQLite3-type environment, I instructed the agent to bifurcate its approach; starting at opposite ends and meeting in the middle for data retrieval latency and making sure both coding modules don’t concurrently overlap and spaghett-ify code. When the model detected a slow query bottleneck, the Lab automatically initiated a sequence:

1. Diagnosis: It analyzed the SQL query structure and the indexing rules.

2. Research: It leveraged Deep Research to pull the latest best practices for PostgreSQL query optimization.

3. Correction: It generated new, optimized indexing code and executed a test deployment.

4. Verification: The Lab then reported a 28% reduction in query latency and updated the codex-beta repository documentation automatically.

Image of Self-Evolved Codebase:

Imagine a dynamic dashboard displaying the code health. On the left, the original, messy version; on the right, the clean, self-augmented version. The core features now showcase evolving functionality with a codebase complete with enhanced augmentations built completely agentically, validated by the Perplexity Lab output in the assets tab.

The Takeaway: This isn't just theory. Agents powered by Perplexity's deep research capabilities are capable of running complex multi-step simulations and applying reasoning to autonomously solve problems—whether it’s optimizing a financial tool or fixing code.

In case you missed it…💸 Here's a reprint of the low-code blueprint that Perplexity Comet helped me uncover and implement and how much money it saved me:

1. Notion as the Universe's Core: I migrated all my tasks (from Google Tasks), projects (from ClickUp), client information (from HoneyBook), and notes/documents (from Notability and Goodnotes) into Notion databases. Notion's customizable templates made this surprisingly easy, transforming complex data into actionable clarity. Notion AI became my constant companion, summarizing client briefs, drafting project updates, generating content ideas, and refining notes directly within my consolidated workspace. Notion offers a unified workspace, combining notes, tasks, and more, and is collaborative in real-time.

2. Notion Calendar: My New Time-Travel Hub: This replaced Fantastical entirely. With native integrations, all my existing calendar events synced, and I could manage my schedule, tasks, and project deadlines from one unified view within Notion.

3. Zapier: The Interstellar Bridge: This is where the low-code magic truly happens. Using Zapier, an automation tool that connects different applications, I built workflows to bridge Notion with my external communications and document management.

    ◦ Automated Client Intake: When a new lead fills out a form (e.g., on my website), Zapier uses an API endpoint to automatically create a new client entry in my Notion CRM, generate a task for follow-up, and even draft an initial welcome email with Notion AI. This replaced HoneyBook's basic CRM features.

    ◦ Email to Task Conversion: A simple Zapier integration now turns starred emails in Gmail into new tasks in Notion, complete with a link to the original email. Goodbye, Google Tasks overload!

    ◦ Automated Meeting Notes: Whenever a meeting is scheduled in Notion Calendar, Zapier creates a linked page in Notion for notes, pre-populating it with agenda points and attendee lists.

    ◦ Consolidated Document & Note Storage: Automated workflows can now take files from various sources (e.g., email attachments, web clippings) and categorize them directly into Notion databases, effectively replacing the scattered storage of Notability and Goodnotes. This simplifies knowledge management and ensures everything is searchable and linked within my Notion ecosystem.

This entire consolidation, largely powered by insights from Perplexity Comet and executed with low-code tools like Zapier, allowed me to cancel subscriptions to Fantastical, Google Tasks (indirectly), ClickUp, HoneyBook, Notability, and Goodnotes. The cost savings are significant, but the operational leverage—the ability to focus on high-value work rather than managing disparate tools—is priceless. It's like going from a scattered constellation of individual stars to a perfectly orbiting digital command center.

Prompt Engineering Gem: The Unsung Hero

Even though this was an "unplanned" detour, the principles of good prompt engineering, which I've discussed often, were crucial. Asking Perplexity Comet precise questions about integrations, cost comparisons, and Notion's specific capabilities, often utilizing structured input, ensured I got actionable, verifiable answers—not just generic advice. My custom instructions in Perplexity Spaces act as a "Prompt Engineering Gem," guiding the AI to deliver highly specific and actionable outputs.

🛸 Comet’s Cosmic Curios

Here’s what’s pinging The Comet’s Tale ☄️ radar as far as the curious uses of AI this week, selecting content not previously featured in the standard recurring sections of the newsletter template:

The Sinister Celebrity Voice Trend: Remember those fun voice generators? They've taken a dark turn. Users are exploiting AI voice cloning tools to generate famous voices for malicious or bizarre purposes. This includes using AI voice simulators to create deepfakes of celebrities saying awful things or even making figures like Emma Watson read Hitler's autobiography. This trend highlights the dark side of accessible generative audio technology, demonstrating how AI can be misused for abuse and propaganda.

The Rise of the Zombie News Site: We’re seeing a new form of digital decay where defunct news sites and beloved blogs (like The Hairpin or iLounge) are being resurrected through cybersquatting, stuffed with low-quality articles created entirely by generative AI. These "zombie" sites sometimes even publish AI-generated content under the names of the original human staff. This showcases a disturbing convergence of AI content generation and digital exploitation.

💬 Cosmic Q&A

Each day, I’ll answer a hand-picked reader’s question.

Q: My team works with highly sensitive data. If we use Perplexity for Government or use Pro features for analysis, how can I be absolutely sure our proprietary information isn't being used to train the underlying AI models? - Sarah, Compliance Officer

A: This is a critical question, Sarah, especially as confidence and trust are prerequisites for making AI-powered search ubiquitous.

When dealing with sensitive data, transparency and security are paramount. Perplexity addresses this challenge in a few key ways:

1. Zero Data Retention for Government: The launch of "Perplexity for Government" specifically promises zero data retention and secure services for federal users, adhering to strict federal data security guidelines. This model is designed to enhance productivity while mitigating the risk of sensitive information exposure.

2. Sonar API Guarantee: For standard enterprise use via the API, the Sonar models are built with a core privacy promise: there is no LLM training on customer data. This assurance is essential for businesses integrating Perplexity's AI into their own products, ensuring that proprietary information remains private.

3. Proactive Data Control: While utilizing general AI tools requires diligence regarding privacy trade-offs, high-level Perplexity offerings like Enterprise Pro for Government provide custom solutions tailored to meet unique needs, offering enterprise-grade security.

The best practice is to understand the specific terms of the service you are using (e.g., Enterprise vs. API vs. Government) and leverage Perplexity's citation-first architecture to verify that all information generated is grounded in public, verifiable sources, not your internal documents.

🔮 What’s Next—A Look Forward

The shift to autonomy means even content generation gets bizarrely efficient.

Tomorrow's edition features the start of our new series: Autonomous Content Engines. We’re going deep on one of the strangest new applications for AI automation, demonstrating how to transform a chaotic input source into a daily, structured output, without human oversight.

We will break down how to build a whacky automation: The "Brainrot-to-Briefing Micro-Podcast Generator."

We'll use Perplexity Labs and a theoretical audio generator (like ElevenLabs, referenced in other content generation contexts) to turn the most absurd viral meme trends or "brainrot" content of the day (like the "Italian Brainrot") into a concise, professional 60-second audio summary, optimized for B2B consumption. It’s an insane yet effective way to prove how Perplexity can structure and normalize any data source, no matter how ridiculous, into actionable content.

🏁 Final Words—Why This Newsletter Exists

We are witnessing the emergence of software that maintains itself. We are shifting from being operators who write code to being architects who design self-improving systems.

The future isn’t being coded by hand; it’s being orchestrated by AI agents.

Don't just watch the future unfold—engineer it! Install Perplexity Comet, start experimenting with Labs and the Connectors, and push your coding projects toward self-evolution today.

I hope with this unexpected twist, we’ve shown that Comet isn’t just a tool; it’s a playground for explorers, tinkerers, and the quietly bold—people who’d rather build the map than wait for directions. It holds real value. Here, every question is an opportunity, every workflow a spark, and every reader a fellow astronaut in this experiment we call “progress.”

Where other newsletters toss you links and wish you luck, we hand you the launch codes and say, “Strap in.”

Curiosity fuels the ride; Comet powers the mission, Perplexity drives the Curiosity. Revolutions!

Keep experimenting, keep sharing, and—if you want tomorrow’s spotlight—hit reply and show us where your story lifts off.

See you in tomorrow’s astral adventure!

And yes, it’s STILL going…how much agentic functionality can it pack in? 🫨

-Comet on! ☄️ 💫

Chris Dukes
Managing Editor, The Comet's Tale
Founder/CEO, Parallax Analytics
Beta Tester, Perplexity Comet
https://parallax-ai.app
[email protected]

P.S. - Struggling with something specific? Want to see something featured in the Workflow Wizardry Week? Hit reply. If multiple people have the same challenge, it’ll become featured in a future issue.