The internet has a trust problem, and it’s not just because social media feeds are filling up with AI slop. AI-generated text and images are now making their way into job applications, product reviews, and even insurance claims, leaving platforms and users alike scrambling to figure out what’s real. 

A handful of startups have cropped up in the past couple of years to become the “trust layer” the internet needs — including Pangram. The startup recently snapped up $9 million for its AI detection system and landed a partnership with Substack, which is now using Pangram’s tech to show readers which of their favorite authors use AI to write their newsletters. Pangram also recently dropped a new AI image detection tool. 

Watch as Pangram co-founder and CEO Max Spero joins TechCrunch’s Equity podcast to dig into the promise of AI detection tools and where to draw the line between AI assisted and AI generated. 

Subscribe to Equity on YouTube, Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod. 

Though AI has seeped into every step of the job hunt for both applicants and recruiters, it has taken time to catch on in interviews. Voice AI is infamously tricky to pull off, as agents get tripped up by accents or “ums.” Behavior that would be second nature to a human—like not interrupting a candidate—turns into a “very, very difficult engineering problem,” according to Ophir Samson, the head of voice AI at recruitment platform Greenhouse. But recruiters, overwhelmed with the glut of applications, have increasingly turned to voice AI tools to screen candidates, with 63 percent of job seekers reporting they’ve encountered an AI interview, according to Greenhouse.

As recruiters turned to AI, so did applicants. Candidates using AI in the interview process has become so common that AI recruitment startups like Ribbon promise to help identify “overly scripted, AI-assisted, or coached” responses—just like those in Christopher’s experiment.

Now there are rumblings that bot-on-bot interviews aren’t duplicitous but a preview of things to come. In many ways, it’s the “next logical stage,” Mark Monaghan, vice president of organizational development at call center company IQor, tells WIRED.

That doesn’t mean he’s entirely happy about it. In Christopher’s case, Monaghan believes, the use of AI was more than justified.

“If you’re going to send a bot to me,” he says, “I’ll send a bot to you.”

After his fifth interview with no follow-up, Christopher decided to apply for one more Everforth Apex job. Perhaps, Christopher thought, the problem wasn’t AI. Maybe it was him. Maybe if Riley were able to speak to the company’s dream candidate, things would go differently. So he created a fake candidate, whom he called Don Dickner, packing a résumé to the brim with qualifications pulled from an open job posting.

After Christopher submitted the application, Riley immediately reached out. This time, the virtual recruiter spoke with ChatGPT—posing as Dickner—for 23 minutes. They discussed “sustainable operational improvement,” protecting “customer experience under volume pressure,” and refusing to let “tribal knowledge drift.” ChatGPT had an answer for every question, and a personal anecdote for every qualification. As the interview wrapped up, ChatGPT said it had a “few more questions I’d like answered.”

“I appreciate it,” Riley responded. “Upon carefully reviewing your application, we will assess your qualifications for the position. Should you meet our criteria, a recruiter from Everforth Apex will reach out to schedule a brief conversation.”

Christopher never heard from Riley again.

from the this-is-why-we-can’t-have-nice-things dept

Back in June, the Supreme Court ruled 6–3 in National Republican Senatorial Committee v. Federal Election Commission that federal limits on coordinated expenditures by political parties violate the First Amendment, opening the floodgates to a much broader array of political ads funded via no limit of rich assholes and their preferred dark money groups.

In preparation for the ruling, Trump FCC boss Brendan Carr revised FCC “Lowest Unit Charge Requirement” rules to try and make it much cheaper for the GOP to pummel the midterm elections with less-expensive TV ads carried via the nation’s soggy assortment of right wing broadcasters (which are currently petitioning the Trump FCC to approve massive new mergers).

But it hasn’t all been easy going for the GOP, which believes its massive funding advantage ($125 million for the GOP versus a bunch of debt for the mismanaged DNC) would give them a real leg up during the midterms.

For one thing, the Richmond, Virginia-based 4th Circuit Court of Appeals recently sided 2-1 against the FCC, temporarily suspending the FCC’s attempt at discount TV agitprop, and ruling that neither political parties nor joint fundraising committees with non-candidate members are allowed the discounted rates.

But the GOP has already set the wheels in motion to get this all quickly overturned by the Trump-friendly Supreme Court:

“The committees submitted an emergency motion for a stay and asked the 4th Circuit to rule on that motion immediately so they can file a petition to the Supreme Court. “Intervenors respectfully request that the Court rule on this stay motion as soon as possible—whether by expediting or waiving response briefs—to permit Intervenors to seek emergency relief at the Supreme Court,” Republican committees told the court.

The court responded quickly, issuing an order today to deny the Republican committee’s motion and to immediately issue a mandate that can be appealed to the Supreme Court. Republicans will now seek swift action from the Supreme Court in an attempt to overturn the 4th Circuit ruling before the 60-day discount period starts on September 4.”

If the GOP wins, local broadcasters will be forced to offer dodgy dark money groups the same discounts previously reserved directly for candidates, something the FCC’s lone Democrat, Anna Gomez, states will be “unleashing a flood of coordinated campaign money into broadcast advertising, just as the Supreme Court has cleared the way for unlimited coordinated spending between parties and candidates.”

It’s another reminder (as if you needed one) that unless the U.S. Supreme court is radically expanded and reformed in the next few years, corruption is likely to strip the country down to parts and sell it for scrap off the back loading dock.

Filed Under: ads, appeal, brendan carr, fcc, midterms, propaganda, supreme court

Klipsch showed us quite a lot of hardware at CES 2026, but some of the most interesting products were sitting behind closed doors and came with the rather important disclaimer that we were looking at concepts.

The new Klipsch Reference Signature Series was one of them.

Eight months later, Klipsch is bringing Reference Signature to CEDIA Expo 2026 in Denver, and this time attendees will actually get to hear them. The company has also confirmed the six-model lineup, disclosed considerably more about the acoustic design, and provided an initial indication of what the speakers will cost when select models arrive this winter.

If Reference Signature sounds familiar, you have been paying attention.

We covered the early versions at CES in January, where Klipsch described Reference Signature as a more ambitious extension of its modern horn-loaded loudspeaker platform alongside the considerably more experimental Project Apollo.

What we did not see at Audio Advice Live 2026 was a Reference Signature launch. Klipsch used the Raleigh show in August for the global debut of its Reference Premiere III Series, including a rather substantial 7.4.4-channel home theater demonstration with Onkyo electronics.

Different speakers. Different series. There are already enough Klipsch model numbers in circulation without us helping create another family reunion nobody can follow.

Related Reviews:

Six Reference Signature Speakers

The new Reference Signature lineup consists of two floorstanding speakers, two monitor models, one center channel and a multipurpose Dolby Atmos, height and surround speaker.

Klipsch Reference Signature Series

• RS-8301F: Three-way 8-inch floorstanding loudspeaker
• RS-6301F: Three-way 6.5-inch floorstanding loudspeaker
• RS-8301M: Three-way 8-inch monitor loudspeaker
• RS-6301M: Three-way 6.5-inch monitor loudspeaker
• RS-6301C: Three-way 6.5-inch center-channel loudspeaker
• RS-6101SA: Two-way 6.5-inch Dolby Atmos, height and surround loudspeaker

The biggest departure from Reference Premiere is not the cabinetry or another variation on Klipsch’s familiar copper-colored drivers.

It is the dedicated midrange driver.

The floorstanding, monitor and center-channel models are true three-way designs with separate high-frequency, midrange and bass drivers. Klipsch says this allows each driver to handle a narrower portion of the audible spectrum and moves the crossover regions away from the frequencies where human hearing is particularly sensitive.

Paul W. Klipsch famously argued that “the midrange is where we live,” and apparently somebody in Indianapolis has been reading the old man’s notes.

A Different Kind of Klipsch Horn

Reference Signature still relies on horn loading and Klipsch’s titanium-diaphragm Linear Travel Suspension tweeter, but the more interesting development is its new phase-aligned horn architecture.

Klipsch positions the midrange directly within the tweeter horn so that the midrange and high frequencies originate from the same acoustic center. The objective is better coherence through the crossover region and more consistent tonal balance across a wider listening area.

That could prove particularly important in a home theater, where the person who paid for the system is rarely the only person sitting in front of it.

The larger Reference Signature models combine 8-inch Cerametallic bass drivers with a 6.5-inch midrange, while the smaller models pair 6.5-inch bass drivers with a 5.25-inch midrange. The floorstanding speakers use dual woofers, while the monitors use a single bass driver.

Klipsch’s spun-copper Cerametallic cones remain part of the recipe, maintaining some visual and technological continuity with Reference Premiere.

The Center Channel Might Be the Most Important Speaker Here

The RS-6301C deserves additional attention because Klipsch has moved away from the conventional horizontal MTM arrangement commonly used in center-channel loudspeakers.

Instead, Reference Signature uses a vertical driver array designed to reduce the horizontal lobing and comb-filtering issues that can make dialogue and tonal balance change as listeners move away from the center seat.

For a serious multichannel system, that is more than an engineering footnote.

The center channel carries an enormous percentage of movie dialogue and front-stage information, yet it is frequently the loudspeaker most compromised by furniture, screen placement and seating position. Klipsch’s approach is intended to provide clearer dialogue and more consistent timbre across multiple seats while maintaining a closer tonal match with the Reference Signature towers.

We want to hear this one from well off-axis before reaching any conclusions.

Atmos Without Locking Yourself Into Atmos

Klipsch has also avoided making the RS-6101SA a one-trick Dolby Atmos module.

The two-way speaker can sit on top of a Reference Signature tower or monitor and fire upward toward the ceiling, but it can also be wall-mounted for use as a direct-radiating height or rear surround channel.

A rear-panel switch changes the crossover configuration between Atmos and Surround operation.

That flexibility matters if somebody changes rooms, moves from reflected Atmos to direct-mounted height channels, or eventually realizes that drilling holes in drywall is less frightening than explaining another speaker upgrade to their spouse.

Klipsch Dresses Up

Reference Signature also represents a noticeable industrial-design upgrade over Reference Premiere.

Klipsch is using a one-piece BMC front baffle designed to provide a denser and more rigid mounting surface while reducing unwanted cabinet resonance. Rounded cabinet edges are intended to reduce diffraction and improve off-axis performance.

Ebony and Walnut finishes feature horizontal grain running along the curved cabinet profile, and Klipsch has added a decorative leather top panel along with metal carpet spikes and rubber feet.

All models will also provide adjustable tweeter output at +1 dB, 0 dB or -1 dB, giving owners a small degree of room or preference-based tonal adjustment without reaching for DSP.

Dual binding posts for bi-wiring and bi-amping will also be available on select models. None of that transforms loudspeaker performance by itself, but Reference Signature clearly needs to look and feel more expensive if Klipsch expects buyers to move beyond Reference Premiere.

The Bottom Line

Reference Signature should not be confused with the Reference Premiere III Series introduced at Audio Advice Live.

Reference Premiere III remains Klipsch’s broad modern passive loudspeaker platform for both two-channel and home theater systems. Reference Signature pushes further with dedicated three-way designs, more elaborate horn geometry, upgraded cabinetry and a considerably more premium presentation.

It is also not Project Apollo.

Klipsch previewed Apollo alongside Reference Signature at CES as a separate, more ambitious exploration of high-performance two-channel audio. Reference Signature is the more recognizable Klipsch proposition: a complete modern speaker family capable of supporting both serious stereo listening and immersive home theater.

That distinction makes the new series potentially more important commercially.

Klipsch does not need another loudspeaker that exists primarily to prove what its engineers can build when cost and living-room diplomacy are removed from the equation. It already has Heritage products capable of occupying that territory rather convincingly.

Reference Signature has to work in actual homes.

Price & Availability

Klipsch is targeting $1,600 to $3,000 per pair, with final pricing and complete specifications to be announced at launch. Select Reference Signature models are expected to arrive in winter 2026.

With CEDIA now underway, we’ll have a chance to hear the new three-way horn architecture for ourselves in Denver and find out whether Reference Signature delivers on the promise of what Klipsch first showed us at CES.

For more information: Klipsch Reference Signature

Related Reading:

A speed test measures three things in under a minute: how fast you can download data, how fast you can upload it, and how long a signal takes to reach a test server and come back. Run one for free at Speedtest by Ookla or Fast.com, ideally over a wired connection, and you’ll have a real answer before your coffee gets cold.

Quick Take

Before You Test

You don’t need to install anything to check your speed. A browser and about three minutes are enough. Close anything that’s actively downloading or streaming in the background first. Another device mid-download will drag your number down, and the test will show a falsely low reading.

How you test depends on what you’re actually trying to find out:

Run the Test

Once you’ve picked your path above, the actual test takes four short steps.

1. Pick a real test tool. Speedtest by Ookla and Fast.com are both free and need no account. Fast.com pulls its test data from Netflix’s own streaming servers, so it doubles as a check on your real Netflix performance, not just a lab number. Running both catches the rare case where one tool’s test server is having a bad day.
2. Connect the way your path calls for. Plugging in with an Ethernet cable gives the cleanest read on your plan’s real ceiling, according to T-Mobile’s testing guide. If you’re testing Wi-Fi, sit close to the router with as few walls in between as you can manage. Testing cellular means switching Wi-Fi off first.
3. Run the test and let it finish on its own. Tap or click Go. Most tests take 20 to 40 seconds and stop automatically once the numbers stabilize.
4. Repeat it three times, at different times of day. Try once in the morning, once during peak evening hours, and once more the next day. Write down all three sets of numbers before you draw any conclusions.

Reading Your Results

Your result screen shows four numbers, and each one tells you something different about your connection.

Download speed is how fast data moves from the internet to your device. It’s the number that decides how fast a video buffers or a webpage loads.

Upload speed is the reverse: how fast your device sends data out. It matters most for video calls, uploading files to cloud storage, and live streaming to an audience.

Ping, also called latency, is how many milliseconds a signal takes to reach the test server and return. Lower is better. High ping doesn’t slow down downloads, but it makes anything real-time feel laggy even on a fast connection. For online gaming specifically, ping usually matters more than download speed, and switching to a faster DNS server can shave off extra milliseconds.

Jitter is how much your ping bounces around from moment to moment. A connection with low average ping but high jitter still feels choppy on calls, because the delay keeps changing instead of holding steady.

What Speed Do You Actually Need?

The FCC’s household broadband guide sorts household needs into three rough bands, based on how many things you’re doing at once rather than any single app.

These tiers come from the FCC’s household broadband guide, and the agency is upfront that they’re rough estimates, not lab-tested numbers. The bottom row reflects a separate, more recent change: in 2024, the FCC raised its official broadband speed benchmark to 100 Mbps down and 20 Mbps up, four times higher than the old 25/3 Mbps standard.

Verify Your Result Against Your Plan

Compare your download number to what you’re actually paying for. Real-world speeds often run a bit below the advertised number, especially during peak hours when your neighbors are all streaming at once. A gap of 10 to 20% below your plan’s rating is normal and not worth worrying about.

A gap of 50% or more, especially across multiple tests at different times, points to a real problem. It could sit with your equipment, your home wiring, or your provider’s network.

If you run a small business or work from home full time and you keep landing short of what you’re paying for, it may be worth comparing a broadband service built for business use, since those plans are usually built for steadier, more simultaneous demand than a typical home plan.

Troubleshooting Slow or Inconsistent Results

Most bad speed test readings trace back to one of a handful of causes. Work through these in order before you call your provider.

Where This Approach Has Limits

A speed test is a snapshot, not an average. It tells you what your connection did for those 20 to 40 seconds, not what it does at 11pm on a Friday when your whole household is home and active.

It also can’t tell you where a problem actually lives. A low reading could point to your provider’s network, your own router, bad wiring in the walls, or a damaged cable. The test alone won’t say which.

A mesh Wi-Fi system or a set of powerline adapters complicates a clean reading further, since each hop between units adds its own small loss. Satellite and fixed wireless home internet behave differently than cable or fiber, too: higher and more variable ping is normal for those connections, even when the download number looks perfectly fine.

Key Takeaways

FAQ

AI AND ML

Predictive text remains available, but users will have to invite it into their workflow

Microsoft is switching off text predictions by default in Word and Outlook after acknowledging that unwanted suggestions can interrupt rather than assist some users.

The change is coming to Word for Windows, the web, iOS, and Android, as well as classic Outlook for Windows and Outlook for Mac. The feature will remain available but will no longer be enabled by default.

Introduced several years ago, the feature is the computing equivalent of Microsoft  peering over users’ shoulders and suggesting what they might type next. Depending on the user and their workflow, it can be a boon, an irritation, or simply something to ignore. The irksome part was that it was turned on by default.

According to Microsoft, “text predictions helps you write documents and emails more efficiently by suggesting words and phrases as you type. While some people find these suggestions useful, others prefer to write without predictive text. This update gives you more flexibility, allowing you to customize your editing experience and create a workflow that best suits your needs.

“This update helps ensure that predictive text remains available for people who find it helpful, while reducing unwanted interruptions for those who prefer to write without suggestions appearing automatically.”

The Windows behemoth did not explain whether the change will disable text predictions for users who have already enabled them or apply only to new installations and profiles. Users who want the suggestions will still be able to switch them on through the applications’ settings.

Microsoft is not the only company to suggest text as users type. Google, for example, introduced Smart Compose in Gmail in 2018.

Microsoft did not say whether telemetry or user feedback prompted the reversal, but it has elected to make the feature opt-in rather than opt-out.

It is a small change, but perhaps a sign that Microsoft is listening to customers who would rather decide for themselves whether AI belongs in their workflow. ®

Original Nintendo DS handhelds from late 2004 keep showing up in repair piles with hinges that snap and shells that powder at the slightest pressure. Owners who stored one in a drawer for two decades often discover the damage only when they try to swap a dead screen or tighten a loose screw. Metallic paint Nintendo laid over the ABS plastic appears to be reacting with the case itself. Yellow or greenish stains leak out around edges, button wells, and hinge roots. Force the clamshell open and the plastic flakes instead of flexing.

Gold Plastic Syndrome was named after a particular annoyance among Transformers collectors who noticed that gold and bronze swirl components began to shatter from the late 1980s onwards, frequently falling apart without any visible stress. The initial wave of successes comprised late Generation 1 figures, Generation 2 combiners, and Beast Wars toys. This state was differentiated not only by the stress-free manner in which the parts dissolved, but also by the shards’ appearance to leave microscopic fine residue for the collectors to clean up.

Nintendo Switch 2 System

• The next evolution of Nintendo Switch
• One system, three play modes: TV, Tabletop, and Handheld
• Larger, vivid, 7.9” LCD touch screen with support for HDR and up to 120 fps

Hasbro later said that the main reason for this was the large number of metallic flecks mixed into the plastic, which simply did not adhere properly, leaving microscopic gaps that accumulated and grew until the part failed with very no force. Sqwerks noticed the same thing on multiple original DS units and Game Boy Advance SP shells with the same gold finish.

Sqwerks spent years dealing with dozens of original DS consoles and only lately began documenting the awful state of the situation late last August, and the results are telling. A sky blue one still has some flexibility in the hinge, but another identical one that has the syndrome is showing signs of deterioration from the lid being pushed around. The paint appears to be a thin layer over a yellow green cast on the graphite black ones. The worst of them just completely break apart, with sides missing and a shredded ribbon cable as a bonus. The board and screen are usually still functional. The case, not so much. It only takes seconds to figure out why; take a screwdriver to a mid panel on an affected case, and it snaps with light pressure. Repeat the process with an unaffected unit, and everything should be OK.

As far as he can tell, no lab study has ever been able to pinpoint exactly what kind of chemistry is at work here, but it’s clear that the additives used to give off that metal sheen and nice swirl finish don’t work well with ABS plastic in the long run, leaving it brittle and prone to shattering. What accelerates the process even more? Age. Combine that with some heat and a layer of paint, and watch it break. No, Retrobrite won’t assist much because it only changes the colors of the old yellowed plastics. Oils that soften old rubber will also be of little use in this situation.

He believes that purchasing an aftermarket shell and getting rid of the factory one before it chooses to disintegrate on its own would be a better choice for people who own an original Nintendo DS in the same state. Save the original board, displays, digitizer, buttons, and other useful components.
[Source]

The Trusted Reviews Awards are back for 2026, and we can finally reveal the shortlist for the top prizes.

Our mission is to provide our readers with the best possible product reviews. We review thousands of products across multiple categories every year with this in mind. From fridges to phones, ovens to gaming handhelds, we pride ourselves on our in-depth testing and original reporting.

The Trusted Reviews Awards is where we celebrate the very best tech reviewed by our experts. Categories for 2026 include Computing, TV & Audio, Mobile, Homes and Cameras. We’re also handing out a special Sustainability Award with eBay.

While the winners will be announced at our ceremony on October 1st, our shortlisted products have now been announced, and they can be viewed below.

A little over a decade ago, I led the controls design for a first-of-its-kind full digital control system for a U.S. nuclear plant. It was, on paper, a beautiful machine—engineered to run itself the way a modern airliner does, with operators watching over a system that rarely needed them. And we made a decision that, to an efficiency-minded observer, looked backward: We deliberately left manual steps inside sequences the system could execute on its own.

We were solving a specific problem. An operator who only ever supervises automation slowly stops being an operator. The hands go cold. The mental model of what the plant is actually doing gets fuzzy. Then comes the day the automation hands control back. It’s always the worst day, because automation only quits when it’s confused or in trouble. But by then, you have a person in the chair who hasn’t truly operated the thing in years. The manual steps were there to keep the human current. It was inefficient by design, on purpose.

That plant, as it happened, was never built. It was shelved amid the politics and economics that surround nuclear power in this country, for reasons that had nothing to do with the engineering. But the design instinct outlived the project, and I’ve come to believe it’s the most useful idea I can offer to the argument now consuming every boardroom: What happens to human expertise when AI does the work that used to build it?

AI Is Disrupting the Engineering Career Ladder

The data has gotten hard to wave away. A Harvard University working paper covering some 65 million workers at more than 280,000 U.S. firms found that after companies adopted generative AI, junior employment fell roughly 9 percent within six quarters relative to non-adopters, while senior employment kept right on growing. A Stanford analysis of ADP payroll records points the same way: The youngest workers in the most AI-exposed occupations lost ground after late 2022 while their more experienced colleagues held theirs. The Stanford researchers found that the losses concentrate where AI automates the work; where it merely augments, junior employment holds steady or rises.

The causal story is still contested, and honesty requires saying so. Researchers at the New York Fed attribute much of the rise in young-graduate unemployment not to AI but to remote work, arguing that firms are reluctant to hire inexperienced people whom they cannot train and mentor at a distance. But notice what the explanations share. Whether a model is absorbing the formative work or distance is severing the mentorship around it, both describe the same broken mechanism: the apprenticeship channel through which expertise passes from senior to junior. Either way, “entry-level” has quietly come to mean “three years of experience required.”

Strip away the noise and you’re left with one deceptively simple problem: you cannot become a senior engineer without first being a junior one. Expertise is not downloaded. It is earned through failed builds, dead-end debugging sessions, and the “why on earth did that work” moments that a capable AI will now happily spare the newcomer. Spare them enough of those, and you produce a cohort that can supervise a model on paper but never developed the gut sense to know when the model is confidently, catastrophically wrong.

Most of the commentary stops at the diagnosis, or reaches for policy solutions that treat the loss of junior jobs as an economic problem. Yet it’s also an engineering problem, and safety-critical fields have already spent decades learning how to solve it.

My own career started at the sharp end of automation. My first job out of school was verifying and validating the software in the digital jet-engine controller that decides, faster than any pilot could, how a fighter plane’s engine responds. Even then, in the late 1980s, the central tension was visible: The machine outperforms the human in routine cases, but the human is all that stands between the aircraft and disaster in the cases the machine didn’t anticipate. This tension is known as the automation paradox, in which increasingly capable automation gives human operators less practice, while leaving them only the most difficult situations.

Aviation learned, repeatedly and expensively, what happens when human skills atrophy inside that gap. The canonical example is Air France flight 447, which fell into the Atlantic in 2009. The proximate cause was mundane. Iced-over airspeed sensors fed the autopilot bad data, and it did what it is designed to do: it disconnected and handed control of the airplane back to the crew. What followed was not a hardware failure. It was a competence failure. A recoverable situation became an unrecoverable one because the pilots, conditioned by thousands of hours of watching the automation fly, could not read a high-altitude aerodynamic stall and hand-fly their way out of it. The airplane was working. The training the automation had quietly eroded was not.

The industry’s response is instructive, and it’s the same move we made in that nuclear control room. It did not rip out the autopilot. It built deliberate manual practice back in. In 2017 the FAA issued Safety Alert for Operators 17007, “Manual Flight Operations Proficiency,” declaring that “manual flight is the foundation upon which other technical flying skills are built.” The alert formally recognized skill decay as a hazard in its own right. Some airlines amended their procedures to encourage hand-flying both the initial climb and initial descent in benign conditions, knowingly trading a sliver of fuel efficiency to keep the crew’s raw flying skills alive. That trade is the whole point. A perfectly optimized system that produces incompetent operators is not optimized at all. It has simply moved its failure mode somewhere the spreadsheet can’t see it.

Manual Gates Could Preserve Engineering Skills

Put the aviation lesson and the nuclear instinct side by side and they point to one design pattern we now need in AI-augmented work: the deliberate “manual gate.”

A manual gate is a point in a workflow where a human takes the controls, not because it is the fastest way to get the task done, and not only as a safety interlock, but specifically to exercise and preserve a skill that would otherwise decay. The distinguishing feature is that it is chosen. You decide, as a matter of design, which competencies your organization must keep alive in human beings because those are the ones you will need on the bad day. Then you engineer the friction required to keep them warm.

Picture how this might work on a software team that leans on AI for most of its code. The team places a manual gate around the skill it can least afford to lose: debugging. When a defect surfaces in a critical module, the assigned engineer—deliberately, often a junior one—must first reproduce the failure, trace it to root cause, and write an automated test that captures the bug, all with the AI assistant switched off. Only after the engineer commits to a diagnosis does the model come back on, to propose the fix, generate alternatives, and sweep the codebase for similar bugs. The engineer then compares their diagnosis against the model’s. When the two disagree, that’s the design working, surfacing the disagreement before the bad day instead of during it.

This approach reframes the junior engineer entirely. The instinct today is to let AI do the entry-level work because it is faster and cheaper. But some of that work is not overhead to be eliminated. It is the training apparatus of your future senior staff, and you should protect it the way you’d protect any other piece of critical infrastructure. It may not be efficient this quarter, but dismantling it quietly mortgages your capability a decade out.

Why Companies Must Keep Training Junior Engineers

None of this is free, and pretending otherwise would insult the people who have to sign the budgets. A deliberate manual gate is, by construction, less efficient in the near term than full automation. Keeping juniors doing formative work and running the manual sequences costs something now to protect something later.

That’s a hard sell in a market that judges most leaders on quarterly results. A hired executive who carries “unnecessary” humans that AI could replace will hear about it from the board long before the payoff arrives. The math only works for someone insulated from that pressure: a founder with control, a private company, an institution with a genuinely long horizon, or a regulator willing to require workers to demonstrate their skills regularly, as pilots must. Which means the organizations most likely to preserve their own expertise are the ones structurally able to spend short-term margin on long-term capability; everyone else will need that outside push.

So here is the argument, in one line: Deliberate inefficiency is not waste. In safety-critical engineering we have always known it as insurance, and we buy it on purpose. As AI takes over the work where expertise is forged, the smart move is not to resist the automation. It is to keep our hands on the controls by design—so that when the automation fails, as it always eventually does, there is still someone in the chair who knows how to fly.