I am hummn.

Three Years In: What ChatGPT Really Does to the Workplace

view thought #198

Overview

The first study on chat-based AI’s impact in the workplace is out. It may feel like AI has been around for a long time—ChatGPT debuted in 2022 and we’re in 2025—but the effects on staffing are nuanced, and early headlines can be misleading.

What the study found

In short: little to no impact on the workforce. The broader implications may vary by industry and role, but the headline takeaway is that AI has not yet displaced broad headcount at scale.

On-the-ground reality

From my experience across hires and departures, AI is not yet at the point of replacing people en masse. It has, however, reduced the need to contract or outsource certain functions, particularly in data analysis, image generation, and copywriting.

Marketing and copywriting in the AI era

For general marketers, the ability to generate large volumes of Google and Meta ads—often as text—as well as rapid email copy and content, is gradually eroding some traditional roles. If you’re a copywriter today, you’ll likely feel the pinch eventually. This trend also affects outsourcing firms in developing nations that rely on copywriting and SEO work.

Transcription is another area where AI shines; it can transcribe voice recordings with impressive accuracy, reducing or eliminating the need for paid human transcription services.

Key takeaways

  • AI is changing workflows, not instantly replacing humans.
  • Outsourcing and contract work in analytics, writing, and transcription are under pressure.
  • Tools like ChatGPT, Google Ads, and AI-powered transcription (e.g., Otter.ai) can boost productivity but also reshape job roles.
Reading time: 2 minutes

AI as a Workflow Co-Pilot: Elevating Creativity and Productivity

view thought #196

In a rare moment of excitement, I walked into the office the other day and, in a very humble way, almost felt like a god. This isn’t my nature to declare such grand things, but AI has completed so much of my mental workflow and the work that I do.

With a broad technical skill set around programming, databases, and especially automation, I can now operate at a higher level. AI from OpenAI helps me:

  • Process analytics at incredible scale with deep, actionable insights. Analytics at scale
  • Generate design ideas quickly and get UX/UI feedback—often using Figma for rapid iteration.
  • Assess usability and identify friction points in concepts and campaigns.
  • Brainstorm campaign ideas, blog topics, and other creative tasks, especially when I’m low on caffeine or inspiration. AI helps fill the gaps when my brain can’t conjure ideas on its own.

Vibe coding isn’t 100% yet, but with persistence you can build very functional software. And even though creative generation isn’t flawless, we’re getting closer with each iteration.

Bottom line: AI is becoming a reliable co-pilot for my workflow, and the horizon keeps expanding as tools improve. If you’re exploring AI in your own work, start with small pilots and scale from there.

Reading time: 2 minutes

Regulating AI in the Age of Deepfakes: The Case for Provenance

view thought #194

Governments really need to think carefully about how AI has already bolted from the stable.

If the future includes deepfake video and audio that can be produced almost effortlessly, how can regulators stay ahead of these assets?

Key challenges

  • Provenance and tamper-resistance: a tag embedded in media that proves AI origin and cannot be easily removed.
  • Watermarks aren’t foolproof: some providers, like Gemini, add watermarks to generated content, but those marks can be removed with common tools such as Adobe Photoshop, undermining provenance.
  • Regulation vs. self-regulation: governments may lean on AI providers to build controls, but the collaboration is unlikely to be fully effective in practice.

Without reliable methods to distinguish real from AI-generated content, the future could be globally disruptive—far beyond what many sci‑fi films have imagined.

What could help

  • Robust, tamper-evident provenance for media and standardized digital signatures.
  • Transparent, independent oversight and detection capabilities.

For example, consider the ongoing work on cryptographic provenance and standard metadata formats. For more on current AI development, see Google Gemini.

Reading time: 1 minutes

Rethinking Social as a Sales Channel: Focus on Paid, Lifecycle Marketing, and Organic Content

view thought #192

After 20+ years in digital marketing, I’m increasingly convinced that social media as a sales channel is overrated for most brands. It works in a few cases, but in most scenarios it’s not the best use of time or money.

Brands should double down on paid channels, lifecycle marketing, and content that earns organic traffic—whether that comes from AI-assisted content creation or from Google Organic.

Why this shift makes sense

Google organic may be shifting significantly—some say it’s effectively dying for many brands. If you’re evaluating a strategy around TikTok or heavy Facebook/Instagram organic activity, consider reallocating budget toward the areas above.

Specific guidance for where to invest:

  • Invest in paid channels with a clear lifecycle marketing approach.
  • Publish content that reliably generates organic traffic (via AI-assisted creation or search).
  • Be mindful of platforms with declining organic reach and reallocate funds accordingly.

If you’re leaning toward TikTok or heavy Facebook/Instagram organic activity, you might want to reallocate some budget toward the above areas.

Reading time: 1 minutes

AI-Driven Commerce: Will LLMs Start Charging for Product Placements?

view thought #190

OpenAI announced commercial integrations this week with merchant tools. Shopify‘s ChatGPT integrations are in the works, and WooCommerce integrations are already available or becoming available through merchant feeds. This signals the next phase of product marketing in AI-assisted commerce.

Key implications for marketers and brands:

  • Paid placements to surface products in search results within AI-assisted experiences
  • Potential charging models for brands to advertise inside LLM-powered surfaces
  • Trust considerations: how LLMs present products while empowering users to choose what to buy
  • Transparency and disclosure around sponsorships and ranking signals

The coming weeks will be telling as platforms test, refine, and debate the balance between helpful recommendations and paid promotions.

What to watch

  • New integrations and pilots across Shopify, WooCommerce, and other platforms
  • Disclosure norms for sponsored AI results
  • Measures of trust, quality, and user control in AI-driven marketplaces

Learn more about these platforms: Shopify and WooCommerce. OpenAI is enabling these capabilities as part of its AI platform: OpenAI.

Reading time: 1 minutes

The AI Content Boom: Excitement, Realism, and the Authenticity Question

view thought #188

I’m not usually one to forecast the future of technology, but I’m increasingly surprised by the joy and excitement people have about AI and its impact on content creation—from video to audio to text.

There’s a growing thrill around content produced at scale: AI avatars, realistic-looking videos, and the idea of creating media faster than ever. Marketing companies are jumping on this trend.

Recent developments I’ve been watching:

  • Corporate interest in AI-led content ecosystems is accelerating the reach and scale of media production.
  • Marketing agencies are validating new workflows that leverage AI-generated assets for campaigns.
  • There are big moves in the industry—reports that TikTok is receiving renewed investment and may be integrated with cloud services to power a larger media empire.
  • AI tools for video generation, like Sora, are expanding the toolkit for storytellers and marketers alike.

The big question: if everyone can produce content that looks authentic and compelling, but without real storytelling or purpose, will we eventually feel overwhelmed and crave genuine human-created content again?

At some point, could this create a window for authentic creators? And if so, who will be able to tell genuine content from AI-generated content?

Reading time: 1 minutes

Blogging with Edge: A Car-Chat About Finding Your Voice (and SEO)

view thought #186

I had a really interesting chat in the car with the family the other day about blogging. Historically, I’ve started a lot of very successful blogs from an SEO perspective, some that I’ve sold, some that I just made money out of until Google decided to kill them.

In all of them, I’ve always been very well behaved, never really felt like I was my true self. I was presenting myself as a professional who didn’t use curse words or speak about anything that wouldn’t be PG. To be honest, in the real world, I swear a lot and am very outspoken. I would like to think relatively humorous, with a good sense of humor, quick wit, with a salty dose of dry sarcasm.

So for one of the first rare times, I’d probably say fuck it and try and be myself a little bit more in this, seeing as I’m really just writing for myself and as an aid to others. If I can help people, I don’t think it really matters whether I’m saying shit or not.

What I learned about authenticity

Here are a few takeaways from that car chat and my own journey:

  • SEO-driven blogs can perform, be sold, or monetize, but they often lose a writer’s authentic voice.
  • Google and other search engines reward usefulness and honesty, not necessarily polish.
  • Showing your real voice — humor, edge, and authenticity — can connect more deeply with readers.

Moving forward

I want blogging to feel like a real conversation instead of a corporate memo. If I can help people, the style matters less than the substance, and the honesty matters most.

For readers who care about search basics and visibility, I still care about SEO, but I’m choosing to be myself first. You might see more personality in what I write, even if it means taking a few risks. For more context on search, check Google.

Reading time: 2 minutes

OCR with AI: Why Editable PDFs Beat Images for Reliable Text Extraction

view thought #184

I’ve been doing a lot of experimentation lately with AI as a tool for OCR on images and PDFs. One tool I built was a scraper that analyzed the resulting image and produced the text from that image.

Here’s what I learned from testing different image formats and models from OpenAI (including GPT-4 and other OpenAI models):

  • PNG vs JPEG: generally solid OCR, but results vary by model.
  • Consistency matters: to get reliable data from PDFs or images, using editable PDFs dramatically improves accuracy because text can be read directly from the file rather than from pixels.

Why editable PDFs matter

  • Most PDFs you receive in invoicing, statements, creditors, and other services are editable PDFs. When the text is embedded as actual text in the PDF, the OpenAI engine can pull it directly instead of performing OCR on an image—reducing errors and noise.

Practical takeaway if you’re exploring OCR with AI

  • Prefer obtaining the PDF version of documents whenever possible. Even if you supply structured JSON with a few-shot prompt, a direct PDF source yields far more reliable extraction than image-based OCR.

Related tools and sources

For model options and AI capabilities, see OpenAI: OpenAI and the ChatGPT ecosystem: ChatGPT. For creating/editing PDFs to enable better text extraction, consider Adobe Acrobat: Adobe Acrobat.

Reading time: 2 minutes

Vibe Coding, WordPress, Bricks Themes, and the Reality of Building SaaS

view thought #169

Over the last four to five months, I’ve been vibe coding a bunch of SaaS products. Like many vibe projects, they sit at about 80% complete, with the final 20% taking up the majority of the effort. It’s amazing how quickly you can produce complex things, even as new errors pop up—the speed is exhilarating, but that last stretch is where the real work happens.

Recently I had to build a site or two for the business, and I returned to my true and faithful WordPress setup with the Bricks theme. Bricks is refreshing for the level of control it gives over layout and components. Paired with ACF Pro, it’s incredible how fast you can output a well-structured, attractive site that’s easy to maintain long term.

In the short term, vibe coding has been an incredible experience. If I ever launch any of these products, I’ll certainly journal about it. For now, the WordPress + Bricks solution feels the most robust, industry-ready, and production-stable option I can build. That said, other solid systems—Webflow and other long-standing CMSs popular over the last decade—remain reliable bets if you’re aiming to create something solid.

  • Vibe coding delivers speed and power, but the last 10-20% is hard work.
  • Bricks + WordPress + ACF Pro enables quick, maintainable, visually strong sites.
  • Choose the tool that gives you production stability, while not discarding the value of proven platforms like Webflow.
Reading time: 2 minutes

From Voice Memo to Published Post: An AI-Driven Blogging Workflow

view thought #167

This post outlines a practical automation workflow that uses AI to convert a voice memo into a published blog post. The process stitches together Pabbly Connect, OpenAI’s APIs, image generation, and WordPress publishing — all with minimal manual steps.

The Core Pipeline

The idea is simple: capture a voice memo, clean it up with AI, generate a matching image, and publish everything to your blog. The components involved are: Google Drive as the storage for memos, Pabbly Connect as the automation hub, OpenAI for transcription and content cleanup, an image generator API, a service to convert images to PNG, and WordPress for publishing.

Step-by-Step Workflow

  1. Record a quick voice memo on your phone.
  2. Save the memo to a dedicated Google Drive folder.
  3. Pabbly Connect monitors that folder for new files (a trigger).
  4. The new file is sent to OpenAI’s API for transcription, turning speech into readable text.
  5. The transcription is routed back to Pabbly Connect, which passes it to OpenAI with a cleanup prompt. The prompt corrects grammar, removes repetition, and formats the text for a post.
  6. OpenAI returns a structured JSON object containing a title, a body, and a set of tags. The title guides the image generation step.
  7. The title is sent to OpenAI’s image generation API to create a relevant image that matches the post theme.
  8. The image is returned as base64 data, which is then uploaded to a service (HCTI) to convert it into a PNG file.
  9. The PNG image is uploaded to WordPress via the Pabbly Connect connection, and the rest of the post text is posted as the content. The image’s media ID is attached to the post to ensure the image appears with the article.
  10. With these steps complete, a full blog post goes live with minimal manual effort — simply by recording a voice memo.

Why This Works

Key benefits:

  • Speed: From voice memo to live post in minutes.
  • Quality: Automatic grammar cleanup and layout improvements reduce editing time.
  • Consistency: The system generates a matching image based on the title and content.
  • Automation: A single voice memo can trigger the entire publishing workflow without human intervention.
  • You can post thoughts anywhere, and fast. In the car, plane, commuting, in the office etc.

Easiest way to blog ever. This pipeline demonstrates how automation and AI can streamline content creation, enabling creators to publish quickly while maintaining quality and consistency.

Reading time: 2 minutes