But it had one glaring problem: it didn’t work with Home Assistant.

Every other smart device in my house — lights, locks, sensors, switches — lives happily inside Home Assistant, participating in automations, reporting state, and playing nicely with everything else. The Homewerks fan? It sat there on its own little island, reachable only through the Homewerks app or Alexa voice commands. Want to automatically turn on the fan when humidity spikes? Tough luck. Want to dim the light as part of a bedtime routine? Not happening.

So I did what any reasonable person would do. I contacted the manufacturer and asked for help. That went about as well as you’d expect.

The Customer Service Experience

In January 2023, I submitted a request through the Homewerks website asking if they had any plans to add Home Assistant integration, or if they’d be willing to let me look at the source code so I could build one myself. Simple enough, right?

The first response asked if I was referring to “the Alexa home assistant.” I clarified that no, I meant Home Assistant — the open-source home automation platform with almost 2,000 integrations and millions of users. I even included a link.

The next reply informed me that “this product is a Homewerks product that has Alexa technology built in. It is not an Alexa product first. So unfortunately, it cannot be connected to Home Assistant.” They seemed to think I was asking how to pair it, not requesting a feature or offering to build one.

I tried once more, explaining that I was aware it couldn’t currently connect to Home Assistant — that was the whole point of my request. I pointed out that creating a Home Assistant integration would increase the product’s value to millions of potential customers who have passed on it specifically because it doesn’t integrate with their setup.

The final response: “The Alexa fan model # 7148-01-AX was engineered to work with Alexa technology, exclusively, as a standalone product. Unfortunately, it cannot connect to the Apple Home assistant.”

Apple Home assistant.

They thought I was asking about Apple HomeKit.

Three different customer service representatives across four emails, and not one of them understood what I was asking for. To be fair, Home Assistant isn’t a household name outside of the smart home enthusiast community. But the unwillingness to even consider the request — or pass it along to someone technical — was frustrating.

So I decided to build it myself.

Finding the Way In

The 7148-01-AX doesn’t expose a documented API. There’s no developer portal, no SDK, no MQTT broker to connect to. From the outside, it’s a black box that talks to Alexa’s cloud services and the Homewerks mobile app.

But it’s still a device on my local network, and devices on local networks can be observed.

This is where Claude came in. I’d been using Claude for various development projects, and the idea of pointing an AI at raw network traffic to help reverse-engineer a proprietary protocol sounded like exactly the kind of problem it could accelerate.

The first step was reconnaissance. We scanned the device’s open ports and found it listening on TCP port 8899 — a port commonly associated with Linkplay, a Chinese IoT module manufacturer that provides the WiFi and audio chipset for a huge number of smart speakers and IoT devices. We also found UPnP services on ports 49152 and 59152, which confirmed the Linkplay connection. The device’s UPnP description identified itself as manufacturer “Linkplay” with model “MUZO Cobblestone.”

Knowing it was Linkplay-based gave us a starting point. Linkplay devices use a custom binary protocol over TCP for UART passthrough — essentially a way for network commands to talk directly to the device’s microcontroller. The frame format uses a specific four-byte header (0x18 0x96 0x18 0x20), followed by a little-endian payload length, twelve bytes of padding, and then a payload string in the format MCU+PAS+{json}&.

With Claude’s help, I wrote Python scripts to connect to port 8899 and start probing. We tried dozens of different JSON command structures — {"STA": "query"}, {"status": "get"}, {"get": "all"} — watching for any response. Most commands disappeared into the void.

Then we discovered something crucial: sending a property key with an empty string value causes the device to report that property’s current state. Send {"fan_power": ""} and you get back {"fan_power": "OFF"}. Send {"light_power": ""} and you get {"light_power": "ON"}. Send {"percentage": ""} and you get {"percentage": 255} — which told us the device uses a 0-255 brightness range internally rather than the 0-100 that Home Assistant expects.

From there, it was a matter of mapping out the full command vocabulary. We discovered commands for turning the fan on and off, controlling the light, setting brightness, and even interacting with the built-in speaker as a media player. Each discovery built on the last, and Claude helped me iterate through possibilities in minutes that would have taken me hours of manual trial and error.

Building the Integration

With the protocol mapped out, the actual Home Assistant integration came together relatively quickly. The architecture is straightforward:

A persistent TCP connection to port 8899 listens for state changes pushed by the device (when someone uses the wall panel or Alexa voice commands), while also providing the ability to send commands back. The integration exposes three entity types in Home Assistant: a fan entity for the exhaust fan, a light entity with brightness control, and a media player entity for the Bluetooth speaker.

The tricky parts were the ones you don’t think about until they bite you. The device only allows one TCP connection at a time, so the integration has to be careful about connection management. State updates from the device sometimes arrive as multiple JSON frames concatenated in a single TCP read, so the parser has to handle splitting those apart. And if the connection drops — which happens, because IoT devices on WiFi are flaky — the integration needs to reconnect automatically with exponential backoff rather than dying silently and requiring a full Home Assistant restart.

That last problem — silent connection death — was one we didn’t catch until the integration had been running for a while. The fan would work fine for days, then suddenly stop responding to commands and stop reflecting state changes. Restarting Home Assistant fixed it every time, which is the classic symptom of a lost network connection with no reconnection logic.

The fix involved adding connection health monitoring (a keepalive check if no data arrives for three minutes), automatic reconnection with backoff from one second up to sixty seconds, entity availability tracking so Home Assistant shows the fan as unavailable rather than just frozen in its last known state, and an initial state query on every connect and reconnect so entities start with accurate values instead of defaulting to off.

What Would Have Taken Weeks Took Hours

I want to be honest about Claude’s role in this project, because I think it illustrates something important about where AI assistance is genuinely transformative versus where it’s just convenient.

The protocol reverse-engineering phase — scanning ports, probing command structures, interpreting binary frame formats, iterating through possible JSON payloads — is the kind of work that traditionally takes a dedicated reverse engineer dozens to hundreds of hours. You’re staring at hex dumps, cross-referencing with known protocols, writing and rewriting test scripts, and doing a lot of educated guessing. It’s fascinating work if you have the time, but I have a full-time job and a nonprofit to run.

With Claude, that phase took a few short hours instead of weeks. I could describe what I was seeing on the wire, and Claude would suggest what it might mean based on its knowledge of Linkplay protocols, IoT communication patterns, and binary frame formats. When a probe came back with an unexpected response, we could immediately adjust our approach. When we hit a dead end, we could pivot to a completely different strategy without the cognitive overhead of context-switching.

The integration code itself was also dramatically faster to write. Claude understands Home Assistant’s integration architecture — config flows, entity platforms, the coordinator pattern, async patterns — and could generate working scaffolding that I then refined and tested against the actual device. The back-and-forth of “try this, it didn’t work, here’s what happened” was natural and productive.

But the key insight is this: Claude didn’t replace me. It accelerated me. I still had to make every architectural decision, test every change against the real hardware, debug the weird edge cases that only show up after days of runtime, and maintain the project going forward. The AI made the tedious parts fast and the hard parts approachable, but it couldn’t have done any of this without someone who understood the goal and could evaluate whether the results were actually correct.

The Result

The Homewerks Smart Fan Integration is available on GitHub as a HACS custom component. It provides local control of the 7148-01-AX’s fan, LED light, and speaker — no cloud services required, no Alexa dependency for basic operations.

My bathroom fan now participates in humidity-based automations. The light dims as part of my evening routine. The speaker shows up as a media player in Home Assistant’s media controls. And it all works reliably, reconnecting automatically when the WiFi hiccups, reporting accurate state even when controlled from the wall panel.

It’s the integration Homewerks wouldn’t build, built by the customer they wouldn’t listen to, reverse-engineered from the protocol they wouldn’t document, and developed in hours instead of months thanks to an AI that actually understood the question I was asking.

What’s Next

The integration works. But “works” and “done” are two very different things.

The day after publishing the initial release, I started using it in earnest — and the cracks appeared almost immediately. The fan entity would show the correct state for a day or two, then quietly stop updating. Home Assistant would still show the fan as “off” while it was roaring away above me in a steam-filled bathroom. The TCP connection to the device had silently died, and without reconnection logic, the integration was deaf to any state changes until I restarted Home Assistant entirely. That became Issue #3.

Fixing it meant going back to the protocol and making a discovery we’d missed the first time around: the device doesn’t respond to status query commands in any documented format. No {"status": "get"}, no {"STA": "query"}. Nothing. But Claude and I found that sending a property key with an empty string value — {"fan_power": ""} — causes the device to report that property’s current state. It was an undocumented quirk hiding in plain sight. We also discovered that the device sends multiple JSON frames concatenated in a single TCP read, which our parser had been silently dropping. And that the device reports brightness on a 0-255 scale while Home Assistant expects 0-100, so half the brightness values were wrong. The fix added automatic reconnection with exponential backoff, periodic polling as a safety net, connection health monitoring, and proper state queries on every connect. Version 1.2.1 turned a fragile proof of concept into something I could actually trust.

Then came Issue #1 — the one that bit us during our own debugging session. Claude and I were trying to probe the device for protocol details, connecting Python scripts to port 8899, and getting nothing back. The device wasn’t even responding to pings. Turns out the fan had gotten a new IP address from my router, and the integration was still trying to talk to the old one. Issue #1 demonstrated itself while we were trying to fix Issue #3. The irony wasn’t lost on me. We probed the device’s UPnP description and found it carries a unique device identifier — a UUID that stays constant even when the IP changes. The plan is to use that UUID to scan the local network when the connection fails, so even if the IP changes, the integration can find the fan again on its own. No more deleting and re-adding the integration every time your router decides to shuffle addresses.

And then there was Issue #2 — a user asked if the microphone mute could be toggled through the integration. Unfortunately, the wall panel connects to the fan unit through standard romex wiring with no data bus, and the mute signal never surfaces on the TCP protocol. That one was a dead end, but fun to investigate as I sat in the bathroom pressing the mute button like a detective desperately buzzing a suspect’s apartment, while Claude staked out the packet capture waiting for a signal that never came. All of these fixes and features probably took less combined time than it took to write this article, thanks to Claude.

People complain about AI a lot lately, and I get it. There are legitimate concerns about the energy consumption and water usage required to run these massive data centers. And when the most visible use case is your social media feed full of “Create a caricature of me and my job based on everything you know about me” posts, it’s easy to look at that and wonder if all that energy and water is worth it. But AI can also be used to build powerful tools that provide long-lasting value — tools that add real functionality to hardware you’ve already paid for, that solve problems the manufacturer won’t solve, and that turn a closed ecosystem into an open one. A generated selfie disappears from your feed in a day. A working Home Assistant integration makes your home smarter for years.

If you have a Homewerks 7148-01-AX and run Home Assistant, check out the GitHub repo. And if you work at Homewerks and you’re reading this — it’s Home Assistant. Not Apple Home assistant. Not the Alexa home assistant. Home Assistant. You should check it out. Your customers are asking for it.

The honest answer sits in the middle.

Before getting into whether C.R.E.A.T.E. is useful or useless, it helps to be clear about what it actually is and where it came from.

C.R.E.A.T.E. is a prompting framework developed by David Birss. Birss is a longtime marketing strategist, author and consultant who has spent decades teaching structured thinking to businesses and educators. He asks users to spell out a character, rules, audience, tone and expected output before asking an AI to write anything. In theory, it forces clarity. In practice, it is usually presented as a repeatable template you can paste into ChatGPT to get “better” results. Like many teaching frameworks, its clarity and structure make it easy to formalize, share and promote.

The Hype Around C.R.E.A.T.E.

Over the past year, YouTubers and TikTokers have latched onto C.R.E.A.T.E. hard. Video after video pitches it as the missing secret to perfect AI output. The tone is often breathless. Use this framework. Never prompt the same way again. Watch your results instantly improve. Some creators talk about it like it is the second coming of ChatGPT-esus.

That is where the hype starts to show.

What makes C.R.E.A.T.E. harder to dismiss is that it has moved well beyond social media. Universities and professional programs now teach it as foundational knowledge. Georgia Tech’s Innovation & Entrepreneurship Institute has published guidance on prompt engineering built around C.R.E.A.T.E., positioning it as a core skill rather than a trick.

NYU’s School of Professional Studies has done the same, framing C.R.E.A.T.E. as a structured method students should learn early when working with large language models. Both programs emphasize that prompt engineering is a critical skill for the future workforce.

When college courses and certificates treat a framework as table stakes, it is worth taking seriously, even if the marketing around it feels inflated.

Frameworks like C.R.E.A.T.E. are easy to sell because they look simple and authoritative. They give people something concrete to hold onto in a space that still feels fuzzy and fast moving. Slap an acronym on common sense and it feels like a system. Systems feel powerful.

What C.R.E.A.T.E. Actually Does

C.R.E.A.T.E. does not make an AI model smarter. It does not unlock features or improve reasoning. What it does is reduce ambiguity.

By forcing you to define a context, rules and output expectations up front, C.R.E.A.T.E. limits the range of possible responses. That matters because large language models respond directly to constraints. The tighter the boundaries, the more predictable the output.

In practice, C.R.E.A.T.E. acts as a checklist. Audience, tone, formatting, do and do not rules. When those are clearly stated, the model spends less effort guessing what you want and more effort producing usable text.

Why It Can Feel Like Snake Oil

C.R.E.A.T.E. is often marketed as if the acronym itself has power. It does not. The model does not recognize C.R.E.A.T.E. as a concept. It only responds to the instructions inside it.

A skilled editor could write an equally effective prompt without ever naming a framework. If you strip away the label and keep the rules, the results stay the same. That is where the hype creeps in. The value comes from the discipline of writing constraints, not from the branding.

Where C.R.E.A.T.E. Genuinely Helps

C.R.E.A.T.E. earns its keep when consistency matters.

It works well for repeatable content like theatre listings, newsletters, summaries or product descriptions. It helps when multiple people reuse the same prompt. It also helps when future you needs to get the same output six months from now and does not want to reverse engineer your intent.

In editorial workflows, C.R.E.A.T.E. functions like a style guide embedded in the prompt. It enforces things humans often forget. No calls to action. Fixed credit formats. Specific language rules. Consistent structure. Those guardrails save time and reduce revision cycles.

Where It Adds Little Value

C.R.E.A.T.E. is unnecessary for one off writing. It also slows things down if the rules change constantly. If you already know how to specify constraints cleanly, the framework adds little beyond documentation.

It is also a poor fit for exploratory writing where discovery matters more than uniformity. Too many rules can flatten voice and limit useful surprises.

The Real Takeaway

C.R.E.A.T.E. is not snake oil, but it is not special either. It is a packaged set of best practices that academia and social media have both embraced for different reasons. Its effectiveness depends entirely on how carefully the rules are written and how consistently they are applied.

If you treat C.R.E.A.T.E. like a shortcut, it disappoints. If you treat it like an editorial contract, it works exactly as advertised.

The difference is not the framework. The difference is whether you are thinking like an editor or just typing and hoping for the best.

People frequently approach AI expecting instant results with minimal effort. But working effectively with AI isn’t a matter of clicking a button and walking away; it requires a collaborative effort, thoughtful prompting and iterative refinement.

Working with AI is a lot like onboarding the world’s smartest intern. At first, it sounds like a dream: an overachiever who never gets tired, is eager to help and has access to nearly limitless information. But then you realize they need detailed instructions for even the simplest job. Artificial intelligence is a lot like that intern — full of promise, capable of greatness but completely dependent on the quality of the guidance it receives.

AI expert Daniel Shorstein described a common scenario in a recent LinkedIn post: “People tell me they tried using ChatGPT for something and it didn’t work. So I asked them whether they tried a follow-up prompt or changing the prompt, and usually the answer is no.” Shorstein hit the nail on the head: “Working with AI is similar to working with a human; if you don’t give them enough context or instructions, you won’t like the results.”

People attempt a task once with AI, become frustrated by subpar results and give up immediately. Rarely do they invest time in refining their prompts or providing additional context. But, as Shorstein rightly points out, effective AI usage demands careful and clear instructions.

The misunderstanding goes deeper. Many people still approach AI as if it were just a smarter search engine: type in a query, get instant results and move on. But AI is fundamentally different. It demands clarity and specificity because, unlike a seasoned coworker, it possesses zero intuitive context. Without clear instructions, it simply guesses, often missing the mark.

This was sharply illustrated by leading AI strategist and former AWS head of machine learning startups Allie K. Miller, who shared an eye-opening Zoom call with a celebrity client struggling with AI-generated product descriptions. Miller quickly diagnosed the issue, noting that the employee responsible had barely provided context: “Turns out the employee gave it ONE example. Described the company in TWO sentences. Didn’t even describe the audience. Used her free ChatGPT account.”

And there’s the rub. Most of us treat AI as a quick fix rather than a collaborative partner. Just as we wouldn’t hand a new employee a vague task and walk away, we shouldn’t expect optimal results from a simple prompt entered into a free AI platform. Effective use demands repeated refinement, experimentation and an iterative, conversational approach.

To unlock AI’s full potential, users must stop seeing it as a shortcut and start treating it as the highly capable but utterly context-dependent intern that it is. Only then can we move beyond the false conclusion that “AI doesn’t work” and instead leverage it as a transformational partner.

Ultimately, it’s not AI that’s letting us down. It’s our own misunderstanding of how to work alongside it. The potential is boundless, but only if we invest in understanding the tools we have and learning to use them with the depth and diligence they demand.

Miller didn’t hold back: “I told her flat-out: that’s not the AI’s fault. That’s the employee’s, and maybe even hers for not knowing how to fix it.” To illustrate her point, Miller showcased an 18-page standard operating procedure tailored for AI writing in her own voice and demonstrated how multiple specialized AI tools could be employed simultaneously. The response? The client was floored by the capabilities she’d overlooked.

“A lot of people think the AI ‘doesn’t work.’ IT WORKS. You’re just using it like Google,” Miller emphasized. AI requires users to “benchmark your inputs and literally overwhelm the model with clarity and specificity and intent.”

Take a recent situation I encountered, which shows exactly why this collaborative approach matters. I faced a complex problem on one of my web servers — a situation that I knew from experience would usually take several hours to resolve. Instead, I spent an intense 30 minutes collaborating with ChatGPT. The session involved detailed troubleshooting, several failed attempts, adjustments and clarifications. But the collaborative effort dramatically shortened the process. What would have normally taken hours was solved swiftly through careful, iterative prompting and active collaboration.

Investing the time to properly use AI isn’t always about speed, either. Often, AI enables significantly better results within the same amount of time. Tasks that used to take four hours can still take four hours, but the quality or scope of the finished product far surpasses what could have been achieved alone.

To effectively leverage AI, consider the following tangible steps:

• Start small and iterative: Begin with simple, clear prompts. Review the output, adjust your instructions and progressively add complexity.

• Provide detailed context: Explain your goal, audience, style and specific needs. AI thrives on clarity and specificity.

• Ask questions: Use the AI to get better results from itself! Tell the AI what you’re trying to achieve and ask it what it needs to know to help you. This can lead to more refined and useful outputs. If the output isn’t what you expected, ask it to clarify or expand on certain points. This iterative questioning can lead to much more refined results.

• Brainstorm with AI: Use it as a sounding board for ideas. Ask it to generate multiple options or a variety of styles, which can help you refine your own thinking.

• Collaborate actively: Treat the AI interaction as a conversation. Ask follow-up questions, refine your prompts and clarify when outputs aren’t matching expectations.

• Be honest: It might sound odd, but being honest with the AI about your opinion of its output can help it learn and adjust. If you don’t like what it produced, tell it why and ask for a different approach. If you do like it, explain what you found useful. This feedback loop can significantly improve the quality of future interactions.

• Experiment across platforms: Different AI tools excel at different tasks. Be open to trying multiple platforms to find the best fit for your needs.

• Benchmark your inputs: Regularly assess the quality of your prompts and the AI’s responses. Adjust your approach based on what works best.

• Learn from examples: Study successful AI interactions and prompts shared by others. This can provide valuable insights into effective strategies and techniques.

Ultimately, the effort invested in learning to collaborate effectively with AI is profoundly rewarding. Instead of treating it like the new kid you shove in the corner with busywork, we need to recognize the partnership and responsibility involved. AI, like any talented intern, will only reach its full potential if we take the time to train, guide and collaborate with it. When we do, the payoff can be enormous: productivity and quality on a level we simply can’t reach alone.

But the group that should be sweating the most? Middle managers. And most of them haven’t even looked up from their meeting invites long enough to notice.

Not because AI is going to eliminate all middle management — but because it’s going to create a brutal divide between the managers who figure out how to work with AI, and those who don’t.

That’s not just a provocative jab at corporate bureaucracy. It’s a data-backed, pattern-recognized forecast. If you’ve spent any time watching how AI is being adopted across industries, the writing is on the digital whiteboard. The most efficient, AI-literate managers are about to manage more people, more projects and more workflows. The rest will be quietly phased out.

Management, especially the middle layer of it, is increasingly about coordination, delegation, documentation, reporting, approvals and status updates — precisely the sort of tasks large language models, workflow bots and AI assistants are already excellent at replicating, and in many cases, improving.

Let’s be clear: Not all managers are in trouble. The ones who lead with vision, develop talent, build culture and resolve conflict are doing the kind of work AI still struggles with — relational, human, deeply contextual. But that’s not what most middle managers are paid for.

A 2023 McKinsey report on middle management productivity found that middle managers spend nearly half their time on nonmanagerial tasks — things like administrative busywork and individual contributor assignments — while devoting less than a third of their time to people management. Worse, they’re spending less than a quarter of their time on strategy, despite identifying it as one of the areas where they deliver the most value. In many cases, they’re not being mismanaged — they’re being misused. And the misalignment between what they’re doing and what they should be doing is exactly where AI-powered managers will pull ahead.

Already, tools like Microsoft 365 Copilot, Slack GPT and Notion AI are quietly absorbing many of the managerial rituals that once required a human: writing recaps, summarizing meetings, assigning tasks, tracking OKRs and pinging team members for updates. With a few more integrations, it’s easy to imagine a future where that entire Monday standup is replaced by an asynchronous report assembled by a machine and reviewed by one manager — not four.

And here’s where the change gets real: the manager who understands how to build, train and trust that AI system becomes more efficient. More scalable. More valuable. They can take on two teams instead of one. Or launch a new initiative without hiring another lead. They’re not being replaced by AI — they’re being amplified by it.

Meanwhile, their peers who resist these tools — out of fear, pride or inertia — fall behind. Not because they’re bad at their jobs, but because they’re suddenly doing less with more. Their performance metrics slip. Their teams grow frustrated. Their promotions stall. Eventually, their role starts to look redundant. Not to the algorithm — but to the organization.

So let’s stop saying that AI is going to replace middle managers. It’s not. But managers who use AI are going to replace the ones who don’t.

This dynamic raises an uncomfortable question for companies: What happens when the people leading your AI rollout don’t realize they’re automating themselves out of a job? Many middle managers are the ones evaluating and approving these tools — without recognizing that they’re slowly rendering layers of their own role obsolete.

To be fair, not every manager will be cut. Just as factory foremen adapted to automation on the floor, many will adapt to managing AI-driven workflows. But if we don’t start being honest about who benefits and who gets squeezed, we’ll miss the opportunity to reimagine management — instead of just downsizing it.

AI isn’t here to flatten the org chart. It’s here to reshape it. And the ones left standing will be those who knew how to feed the machine — not fight it.

But a human — who knows how to use AI better than you — will.

This distinction matters — not just semantically, but existentially. It reshapes how we should think about work, power and responsibility in the dawning era of machine intelligence. Because while science fiction has long warned us about sentient androids rising up to replace us, the real threat is much more mundane: a peer, a competitor, a go-getter who sees AI not as a menace but as a multiplier.

The narrative of automation-driven obsolescence has haunted workers for centuries, from the Luddites of the Industrial Revolution to the factory laborers displaced by robots in the 20th century. AI, especially the generative kind, has added new fuel to those fears — and understandably so. Chatbots write emails. Image models generate ad campaigns. Algorithms draft legal contracts, summarize meetings, code apps, diagnose illnesses, compose music and mimic human voices with uncanny fluency.

But here’s the thing: they’re not doing it alone.

For all its speed and statistical brilliance, AI lacks context, judgment, taste, tact, ethics and accountability. It doesn’t know your brand guidelines, your client’s temperament, your boss’s unwritten expectations or the delicate power dynamics of a boardroom. It doesn’t understand what’s politically risky, socially inappropriate or just bad timing. You do.

That’s why the winning formula in today’s economy is not AI alone — it’s AI plus you. Or, more precisely, AI plus the version of you that knows how to wield it.

We’re entering what economists call a “complementarity era,” where the most valuable workers are not the fastest or even the most skilled in a narrow sense, but the most adaptive. The most curious. The ones who know how to ask better questions, frame better prompts, fact-check outputs and shape AI-generated drafts into something smart, nuanced and useful. In short: the people who know how to work with AI, not against it.

This has already begun. Marketers use AI to test campaign variations in minutes that once took days. Lawyers feed legal models entire case files to spot precedents. Teachers generate individualized lesson plans. Real estate agents automate listings, descriptions and pricing insights. Coders use AI pair programmers to ship features faster. The gains are not magical; they’re practical — and they compound quickly.

What this means, uncomfortably, is that AI is becoming the new Excel. It’s not replacing the job — it’s becoming a prerequisite for doing it well.

If you ignore this shift, your job may well become vulnerable. Not because the algorithm wants it — but because the person next to you is using the algorithm to do more, faster, better. They’re not superhuman. They’re just augmented.

This dynamic raises urgent questions for employers, educators and newsroom leaders. Are we training students not just in facts and formulas, but in how to think with machines? Are companies giving workers time and support to learn new tools — or quietly rewarding those who race ahead on their own? Are editors giving journalists the freedom to explore AI ethically and responsibly — and holding them accountable when mistakes happen, to prevent future mishaps like the Chicago Sun-Times’ AI-generated “Best of Summer” reading list? Are we still measuring productivity in old terms, or rethinking output through a hybrid human-machine lens?

Perhaps most crucially, are we being honest about what’s happening? Because the narrative of “AI will take your job” is passive, and thus disempowering. It makes the future feel like a storm you can’t escape. But “someone using AI will take your job” is active. It’s a call to action.

So let’s stop waiting for the singularity. Let’s start preparing for the coworker who already figured out how to use GPT to do in one hour what used to take you three. That person didn’t get replaced. They leveled up.

Now it’s your turn.

Whether you’re using a computer, phone or tablet, here’s how to clear your cache and cookies in every major browser.

What Are Cache and Cookies?

• Cache: Temporary files stored by your browser to help websites load faster.
• Cookies: Small files websites use to remember your preferences, logins and activity.

Over time, these files can build up, cause errors or display outdated content. Here’s how to clear them.

How to Clear Cache and Cookies on Desktop Browsers

Google Chrome (Windows/Mac)

1. Click the three-dot menu (top-right).
2. Go to Settings → Privacy and security.
3. Click Clear browsing data.
4. Choose Cookies and other site data and Cached images and files.
5. Select a time range (e.g., “All time”).
6. Click Clear data.

Mozilla Firefox

1. Click the menu button (≡).
2. Go to Settings → Privacy & Security.
3. Scroll to Cookies and Site Data.
4. Click Clear Data.
5. Check both boxes and click Clear.

Microsoft Edge

1. Click the three-dot menu.
2. Choose Settings → Privacy, search, and services.
3. Under Clear browsing data, click Choose what to clear.
4. Select Cached images and files and Cookies and other site data.
5. Click Clear now.

Safari (Mac)

1. Click Safari in the top menu.
2. Select Settings → Privacy.
3. Click Manage Website Data.
4. Click Remove All → Remove Now.

How to Clear Cache and Cookies on Mobile Devices

iPhone/iPad (Safari)

1. Go to Settings → Safari.
2. Tap Clear History and Website Data.
3. Confirm when prompted.

Android & iPhone (Chrome)

1. Open Chrome and tap the three-dot menu.
2. Go to History → Delete browsing data.
3. Tap Browsing datea and then make sure *Cookies, site data Cached images and files are checked.
4. Tap Confirm and then tap Delete data.

Android (Samsung Internet)

1. Open Samsung Internet.
2. Tap the menu → Settings → Personal data.
3. Tap Delete browsing data.
4. Select Cache and Cookies and site data, then tap Delete.

How Often Should You Clear Cache and Cookies?

• Monthly for regular maintenance
• Immediately if a website isn’t loading properly
• Periodically if you’re concerned about tracking or storage space

Clearing cache and cookies helps ensure your browser runs smoothly and securely.

Final Tips

• You’ll be signed out of most sites after clearing cookies.
• Bookmarks and saved passwords usually remain untouched — but double-check your browser settings to be sure.

In light of these sobering statistics, I felt compelled to investigate products that have been developed to combat these horrors. I began my exploration with Elepho, a company known for its dedication to child safety. Their product, the eClip Baby Car Seat Alarm, caught my attention, promising to be a tech-savvy solution to avoid forgetting the most precious cargo in the backseat.

The Elepho eClip promises to be a tech-savvy solution to avoid forgetting the most precious cargo in the backseat | 📷 credit: Elepho

The eClip is a small, Bluetooth-enabled device that attaches to a car seat or seatbelt. It’s designed to alert parents and caregivers if a child is left in the car, or if the temperature in the car reaches dangerous levels. The device pairs with a smartphone app that provides real-time alerts and notifications.

Upon arrival, the eClip’s packaging channeled an Apple-esque aesthetic, a nod towards elegance and simplicity. In stark contrast, the device itself felt like a lightweight, inexpensive plastic toy.

The eClip is designed to be fastened onto a baby’s car seat strap, the seatbelt, or even onto a diaper bag via an included lime-green accessory strap. But there’s a snag: the eClip dislodges from the strap all too easily, making it a tempting plaything for your little one — and it’s all too likely to become a forgotten object on the car floor.

The eClip can be used in a variety of ways, but it’s not always easy to keep it in place | 📷 credit: Elepho

I expect an app linked to a child safety product to guide me thoroughly through the setup and ensure that I fully understand how to operate it effectively. However, the only setup directions for the eClip were found on the included paper instructions, which regrettably seemed out of sync with the actual functionality of the device, as I’ll explain later.

The QR code in the paper instructions was easy enough to use, but the web developer in me was a little annoyed at how small the buttons for the App Store and Google Play were on the website. It’s a minor thing, but it foreshadowed a multitude of design problems with the eClip app.

I would spend most of my time on the “Cannot connect with eCLIP” screen of doom | 📷 credit: Ray Hollister

Upon launching the app, I was greeted with a clutter of options and an error message. The app stubbornly refused to connect with the eClip. After much trial and error, I managed to pair the device with my phone, but the user experience was far from intuitive. The initial setup was more of a wrestling match. I was only able to get it to connect to the app through a lot of trial and error using my years of experience with ornery Bluetooth devices.

The testing phase revealed the eClip’s penchant for forgetting its pairing with the phone.

During my first test run, I ventured away from my desk with the eClip app open. I’d almost reached my mailbox (about 40 feet from my desk) when the eClip alert triggered - “eClip connection alert. Have you taken your child out of the car?”

Opting to close the app, I headed back to my office, only to embark on the arduous journey of re-pairing the device. Despite following the printed instructions to a tee, the app remained stubbornly disconnected, offering no guidance on how to proceed. Only when using a method not documented in the instructions was I able to get the device to reconnect, but even that took multiple attempts.

For the second round, I decided to keep my phone in my pocket with the screen off. As I strolled away from the device, the visual alert was absent but the audio alarm rang out clear and strong.

But yet again, my return to the desk was marked by a disconnection between the app and device, forcing me into the tiresome re-pairing ritual once more.

To evaluate the eClip’s temperature alerts, I arranged a series of tests. I began by placing the eClip next to a Maverick Stake wireless bluetooth thermometer on a baking sheet, and slid it into the oven. I set the oven to its lowest setting of 170 degrees Farenheit. The oven’s temperature had already hit 120 degrees Farenheit on the Stake and the internal oven thermometer before the eClip finally showed 95 degrees Farenheit and triggered its alert. Even though the oven heats up faster than a car, the eClip’s delayed response was somewhat disappointing.

Even though the temperature was never going to reach melting temperature, I laid a parchment paper down just in case | 📷 credit: Ray Hollister

Next, I sealed both the Maverick Stake and the eClip inside a mason jar with a metal lid, and positioned it in the sun on my driveway. With the temperature on the porch nearing 100 degrees Farenheit it didn’t take long for temperature in the jar to reach dangerous levels. Here, the eClip was more responsive to temperature changes than in the oven, but it still lagged by several degrees before alerting me to the rising temperature.

Who knew my childhood hobby of trapping bugs in mason jars would one day upgrade to testing safety gadgets? | 📷 credit: Ray Hollister

For the final test, I moved to the colder end of the temperature spectrum, and placed both devices inside a chest freezer. Curiously, after about 10 minutes in the freezer, the eClip registered a low of 46 degrees Farenheit before it finally triggering its “eClip High Temperature Alert,” even though the threshold was set to 55 degrees Farenheit. The Maverick displayed an even lower reading, but as a meat thermometer, it didn’t provide an exact temperature below freezing. The temperature in the freezer was consistently at 0 degrees Farenheit, so the eClip’s reading was off by a significant margin.

The eClip and Maverick Stake chilling with the green beans. | 📷 credit: Ray Hollister

In my testing, the eClip’s temperature monitoring proved to be outright inadequate. The alerts lag significantly, rendering them essentially useless in a real-time dangerous temperature situation. Interestingly, the default ‘off’ setting for the temperature alerts raises some questions. It left me wondering if it was an intentional to divert attention from the device’s less-than-stellar performance.

When it comes to the audible alerts on the eClip, they also leave a lot to be desired. They seem haphazard and oddly inconsistent. The eClip features two distinct voices, one male and one female, for its connection and temperature alerts. Upon testing, the contrast between the two was startling. The female voice for the ‘eClip connection alert. Have you taken your child out of the car?’ provided a clear message, whereas the male counterpart came across as notably softer and less alerting.

The temperature alerts, however, took inconsistency to a whole new level. While the female voice managed to retain some sense of urgency, the male voice recording was downright perplexing. It had an almost leisurely pace and tone, devoid of any alert-like qualities. Not only could you hear audible breaths and clicking sounds indicative of the start and stop of recording, but it also carried a casual storytelling vibe rather than serving as a life-saving alert. The production quality seemed lackluster, almost as if Elepho simply used the raw voiceover files without any professional editing or tuning for the purpose of an alert. The eClip’s alerts, unfortunately, strike as an odd mishmash rather than a well-thought-out, user-friendly feature.

Given the eClip’s retail price of $43.95 (plus $6.99 shipping) on Elepho’s website, it feels overpriced, particularly when compared to the more reliable and feature-rich Apple’s AirTag and the Tile Pro.

In sum, while the eClip’s intentions are commendable, its delivery falls short. Between the lack-luster product build, the finicky app, disappointing temperature alerts and unreliable connectivity, it earns a score of just 3 out of 10 beards.

No device can replace the vigilance of a parent or caregiver, but numerous market solutions aim to help prevent such calamities. If there’s a specific device you’d like reviewed, feel free to reach out via the contact options on my homepage at rayhollister.com.

Keep in mind, the temperature in a parked car can surge by 20 degrees within 10 minutes. Moreover, a child’s body temperature can rise 3-5 times quicker than an adult’s. Even on a pleasant 70-degree day, the interior of a car can heat up to 120 degrees within half an hour. Fatalities can occur if a child’s body temperature reaches 107 degrees.

Never intentionally leave a child alone in the car, not even for a moment. Always double-check the backseat before locking the vehicle, especially if there’s been any disruption to your usual routine.

The service itself looks similar to Twitter but has a familiar feel to Instagram users. It supports text posts of up to 500 characters, as well as photos and videos of up to five minutes. Threads also support reposts — its version of a retweet — and quote posts. As reported by Engadget, ‘posts from Threads can be easily shared to users’ Instagram Story for added visibility.’

Meta claims to have plans to make Threads cabable of supporting ActivityPub, the open-source protocol that powers decentralized services like Mastodon. Meta explained in their announcement that Threads aims to ‘usher in a new era of diverse and interconnected networks’. The company hopes that ‘people using compatible apps will be able to follow and interact with people on Threads without having a Threads account, and vice versa’. This is a significant step forward for the decentralized web, as it will allow users to interact with each other across different platforms.

Threads gained a lot of it’s initial traction quickly by drawing from Instagram’s user base. However, recent data shows that user engagement with Threads has experienced a significant drop since its launch. According to Gizmodo, data from Sensor Tower and Similarweb indicate that ‘daily active users dropped 20% from Saturday while time spent on the platform fell 50% from 20 minutes to 10 minutes.’ Gizmodo also highlighted that Threads is currently unavailable in the EU due to regulatory concerns. Meta is actively blocking EU users from accessing Threads via VPN, further limiting its global reach.

Political scrutiny is another challenge that the new platform faces. According to CNBC, U.S. lawmakers, led by House Judiciary Chair Jim Jordan, R-Ohio, have extended their social media investigation to include Meta’s Threads. Jordan has asked Meta CEO Mark Zuckerberg to hand over documents about content moderation on Threads as part of the panel’s ongoing investigation of tech platforms’ policies and contact with the Biden administration. This is an early indication of the added spotlight Meta’s newest product could bring to the company in Washington.

This evolution in social media is reflected in the opinion of Sriram Krishnan. He suggests that these recent events represent a “fundamental rejection of how the internet and large tech companies have worked for several decades”. Instead of a landscape dominated by a few large companies, we may be entering an era where consumers have more power and rights online. He believes this could be the beginning of a more decentralized internet where no centralized gatekeeper can delete a user’s account or data, and users can take their audience with them wherever they go. As he states, “Technological breakthroughs and unrest were needed to shake things up, and that has happened. A pessimist might say that this is going to lead to chaos and challenges. As an optimist who invests in technology entrepreneurs for a living, I believe we are in for an age of major innovation, with all of us having more options and say in how things are run online”.

While it’s clear that Threads is facing an uphill battle, Meta remains optimistic about the platform’s future. A Meta spokesperson told Gizmodo, ‘While it’s early days, we’re excited about the initial success of Threads, which has surpassed our expectations. We launched the app just over a week ago, and our focus now is on ensuring stable performance, delivering new features, and continuing to improve the experience in the coming months.’ Indeed, the platform has already garnered well over 100 million users.

As it stands, Threads is an interesting experiment in social media. It promises a new way for users to engage with each other, one that builds on the strengths of existing platforms while aiming to avoid their weaknesses. However, only time will tell if it can meet these ambitious goals and become a true contender in the social media landscape.

Will Threads Kill Twitter? - New York Times

Will Threads kill Twitter? - The Week

Meta’s “Twitter killer” app Threads is here - CBS News

Can Threads Be the ‘Twitter Killer’? - Wall Street Journal

How Threads Could Kill Twitter - Wired

Welcome To ‘Threads,’ Facebook’s Twitter Killer - Forbes

Could Threads kill Twitter? (Like, for real?) - Poynter

Side note: I love the hesitance that the editor at Poynter had in writing that headline. It’s like they’re saying, “I don’t know, maybe?” 🤷🏻‍♂️ 🤣

I think it’s more complicated than that. I had the pleasure of discussing Threads and its impact on the social media landscape with my friend, Al Letson this Tuesday on First Coast Connect.

Twitter’s openness was its main attraction, it was powerful and ubiquitous. However, since Elon Musk’s takeover, Twitter has been systematically dismantling the very things that made it unique. The most noticeable change was the limitation on tweet visibility, making it so that you have to be signed into Twitter to get access to any tweets and capping the number of tweets you can see per day to 600. It was allegedly to cut costs for Twitter, but it came across like a temper tantrum. This caused a substantial backlash, and that’s where Threads comes in.

Threads, which wasn’t entirely ready for primetime, was fast-tracked for launch in response to Twitter’s controversy. It launched without much of the functionality one might expect from a social media app, but a lot of it has been patched up since then. However, there are still some sizable gaps, like not being able to compose a thread of posts on an app ironically called Threads.

The defining feature that sets Threads apart from Twitter is its openness. Twitter has an open API that allows the creation of bots and apps, something Musk is trying to clamp down on due to costs. Conversely, the Instagram API is notoriously very limited and requires strict approval to use, resulting in fewer bots and spam. It’s safe to assume Threads will likely follow in Instagram’s footsteps.

So, will Threads be the “Twitter killer?” My answer is no. Twitter’s future lies in Elon Musk’s hands, and any demise will be of his own doing, not due to the emergence of Threads. But Threads does provide an alternative platform for those into microblogging and for those disenchanted with the direction Twitter is heading.

As for the future of social media, I see it as a continually separated world driven by content and users. We will see people migrating to Threads from Twitter, much like a social media exodus. The dream of a unified social media platform, where all apps interact and everyone talks to each other, remains, for now, a dream. Owning a user base is too valuable, and companies are unlikely to relinquish that control.

In the end, we’ll always have a plurality of social media platforms — it’s the nature of the game. It’s a world of competition and choice, much like the competition between CNN and Fox News, or even Coke and Pepsi. With Threads, we’re just adding another choice to the mix, and it’s the users who will ultimately decide its fate.

Finally, I will leave you with this, if do you read one more deep dive on Threads, make it Alexandra Petri’s opinion piece in the Washington Post. She had the guts to say what we all were really thinking last week.

Don’t fret! I’ve got a super-simple solution for you - the TikTok safe-space template, like the one you can snag here on my website at rayhollister.com.

Why a Safe-Space Template is Your TikTok Bestie

TikTok’s vertical video format is like a trendy studio apartment – small but mighty. And just like you’d want to make the most of every corner in your apartment, you want to do the same with your TikTok video space. That’s where the safe-space template comes in!

Clear as Crystal Videos

The safe-space template is like your personal video guide. It ensures that all the important bits of your content – the hilarious punchlines, the killer dance moves, your brand logo – stay visible and clear for your viewers. No more obscured texts or overlaid logos!

More Engagement? Yes, Please!

When your video content is crisp and clear, your viewers are more likely to stick around, give you a thumbs up, and share your content. The safe-space template helps you design videos that will have your audience hooked from the start. More likes, shares, and comments, here we come!

Getting Fun and Creative with the TikTok Safe-Space Template from RayHollister.com

Ready to start the fun? Here’s how to use the TikTok safe-space template from rayhollister.com.

1. Grab the Template: Head over to rayhollister.com/tiktok-safe-area-templates/ and download one or more of the TikTok safe-space template. Easy-peasy!
2. Open it Up: Open your new template using any video editing tool you like. Could be Videoshop, Adobe Premiere Pro, Final Cut Pro X or anything else you fancy.
3. Let the Fun Begin: Time to get creative! Use the template as your guide to position your video’s elements. Make sure all the fun stuff stays within the safe zone marked by the template.
4. Double Check Your Masterpiece: Before you hit the publish button, give your video a final lookover with the template. This will make sure that all your fun and important elements are within the safe area and ready to wow your audience.

So, ready to step up your TikTok game? With the TikTok safe-space template from rayhollister.com, you can say goodbye to obscured video elements and hello to clear, engaging, and viral-ready TikTok content. It’s not just a tool, it’s your secret weapon to being a TikTok superstar. Go forth and create!