Are We Coding Our Flaws into AI? A Deep Dive into NLP
Ready to dive into some mind-bending ideas?
(This episode is AI generated from the LinkedIn Start With AI Newsletter - https://www.linkedin.com/pulse/hour-used-take-three-days-heather-v-masters-graje/ )
Today, we’re tackling a real conundrum that’s shaking up how we think about work: the collapse of time in knowledge work. Imagine cranking out a massive 57-page legal handbook in just one hour—sounds like a sci-fi dream, right?
Well, that’s the reality we’re exploring, and it’s got serious implications for how we bill clients and perceive our own value.
We’re unpacking a fascinating newsletter that argues the traditional hourly billing model is on its last legs as AI tools evolve and take over grunt work faster than you can say "billable hours." Join us as we ponder, and maybe even squirm a little at the thought of how these changes are reshaping the landscape of expertise and pricing in a world that’s moving at lightning speed!
The Details:
In this episode, we’re tackling some hefty realities about the future of work, AI, and the collapsing value of the billable hour.
The core of our discussion stems from a thought-provoking newsletter that’s buzzing in the NLP community. We start with a vivid scenario: a legal professional rewrote a 57-page handbook in just one hour, illustrating a monumental change in how cognitive tasks are performed. This isn’t just about speed; it’s about the fundamental rethinking of how we define and measure value in our work.
Our hosts explore the implications for pricing strategies and the psychological discomfort that arises when clients expect to pay for hours rather than results. We also delve into the urgent call for NLP practitioners to engage with AI’s evolution, shaping its foundational structures before default behaviours set in.
The conversation takes a philosophical turn as we question what happens when we encode our own biases and flaws into these powerful tools. Are we creating a future that mirrors our best qualities, or are we embedding our deepest insecurities into the very fabric of AI?
This episode is a whirlwind of insights that challenge us to rethink our roles in a rapidly changing world.
Chapters:
- 00:00 - Understanding Client Billing Practices
- 01:23 - The Implications of AI in Knowledge Work
- 05:46 - The Impact of AI on Work Efficiency
- 06:51 - The Shift from Time-Based Value to Expertise-Based Value
- 10:55 - Understanding the Transition from NLP to Human-Centric AI
- 15:53 - The Ethical Dilemma of AI Identity Encoding
Takeaways:
- The podcast highlights a significant shift in knowledge work, showcasing how tasks that once took days can now be completed in mere hours thanks to AI, fundamentally altering how we perceive value and pricing.
- We discuss the challenge of moving away from hourly billing, as clients grapple with the psychology of paying for results rather than time, which can lead to discomfort for both service providers and clients alike.
- The author of the LinkedIn newsletter warns that anyone still charging by the hour in 2026 risks their income, as the landscape of cognitive labour is rapidly evolving and requires a new approach to pricing.
- We explore the concept of 'metacognition by design' in AI, where systems like Claude Opus 4.7 are not just faster, but actually capable of self-evaluation, leading to a deeper understanding of what it means to work alongside AI.
- The episode dives into the implications of AI on the apprenticeship model, questioning how we can teach and pass on knowledge when the workflow is increasingly automated and the skills required are changing.
- Lastly, we emphasise the need for NLP practitioners to actively shape AI's foundational psychology, as the current moment presents a unique opportunity to influence how AI interacts with human understanding and decision-making.
Companies mentioned in this episode:
- Claude
- Anthropic
- OpenAI
- Perplexity
Transcript
So imagine billing a client for like three solid days of intense, just grueling legal work, but knowing it actually took you one single hour.
Speaker B:I mean, that is quite the hook, right?
Speaker A:Welcome to the deep dive.
Speaker A:Today we are looking directly at you, the listener.
Speaker B:Yeah, maybe you are prepping for a major strategy meeting today or trying to figure out where your industry is headed.
Speaker A:Or maybe you are just trying to understand the invisible forces that are currently rewriting the rules of how we work.
Speaker B:We have got our hands on a piece of source material today that feels less like a tech update and more like a dispatch from a profound revolution in cognitive labor.
Speaker A:Oh, absolutely.
Speaker A: ,: Speaker A:And it's aimed primarily at NLP practitioners, neuro linguistic programming, along with business leaders and coaches.
Speaker B:But the shockwaves of what this author describes, they spill over into absolutely every knowledge, work, profession on the planet.
Speaker A:It is a phenomenal document, honestly, because it skips right past the usual surface.
Speaker A:The author is dissecting this staggering collapse of time in knowledge work.
Speaker A:Yes.
Speaker A:And diagnosing why the current pricing model you are probably using right now is, you know, fundamentally broken.
Speaker B:Exactly.
Speaker B:We are also going to examine a rapidly closing window of opportunity, one where you can actually shape the foundational logic of how AI processes information.
Speaker A:Okay, let's unpack this because we really need to start right at the beginning.
Speaker A:The author hits us with a very specific, grounded anecdote.
Speaker B:Yeah.
Speaker B:They had to rewrite a 57 page staff handbook because UK compliance legislation had changed.
Speaker A:Anyone who has ever touched compliance or legal documentation knows that is just a brutal assignment.
Speaker B:Oh, it's a total nightmare.
Speaker B:You aren't just changing a few words here and there.
Speaker B:You have to completely restructure the document around new legal lines.
Speaker A:Right.
Speaker A:You have to cross reference clauses and ensure the tone remains professional.
Speaker A:But, you know, legally minding.
Speaker B:Plus you have to map the whole thing to corporate hierarchy.
Speaker B:The author notes that just six months prior, this exact task would have been three solid days of sitting at a.
Speaker A:Desk drinking way too much coffee.
Speaker B:Yeah, exactly.
Speaker B:Just manually grinding through pages.
Speaker A: n this specific week in April: Speaker B:One hour.
Speaker A:One single hour to restructure a 57 page legal compliance handbook.
Speaker A:The author notes that Claude did most of it and they just like steered.
Speaker B:Right.
Speaker B:And the human reflex here, I mean,.
Speaker A:My reflex honestly, sure is to look.
Speaker B:At that one hour timeframe and say, wow, the processors are getting so much faster.
Speaker A:Yeah.
Speaker A:The token generation speed is off the charts.
Speaker A:Right.
Speaker B:But the author insists that is a total misdiagnosis of what is actually happening.
Speaker A:What's fascinating here is the newsletter explicitly points to the release of Claude Opus 4.7, which had dropped just a week prior.
Speaker B:And this AI isn't just a faster typist.
Speaker B:It.
Speaker B:It possesses what the author calls metacognition by design.
Speaker A:Metacognition, literally thinking about thinking.
Speaker A:Yeah, but let's be real for a second.
Speaker A:Is it actually thinking or is this just a secondary verification loop running in the background?
Speaker B:Well, that's a debate.
Speaker B:Right?
Speaker A:Because it is very easy to confuse a relentless stacking of technological leaps like those daily releases from Anthropic, with actual cognitive awareness.
Speaker B:I think we have to look at the functional output rather than the philosophical definition of consciousness here.
Speaker B:Mechanically, Cloud Opus 4.7 is structured to catch its own logical faults.
Speaker A:Right.
Speaker A:During the planning phase.
Speaker B:Exactly.
Speaker B:Before it even begins executing the task, it decides how much computational thinking the specific prompt actually requires.
Speaker A:Wow.
Speaker B:Then it verifies its own work against the initial constraints before it even reports back to the user.
Speaker A:So if we were looking for a way to visualize this, I mean, traditional AI was kind of like swapping a handsaw for a power saw, right?
Speaker B:Sure.
Speaker A:But this new metacognitive layer makes it like the power saw now measures twice before it cuts, realizes the wood is warped, and suggests a completely different angle.
Speaker B:That is a much more accurate way to look at it.
Speaker B:Yeah.
Speaker B:You are no longer just giving a command to a machine that blindly executes it at light speed.
Speaker A:You are collaborating with an agent that evaluates its own methodology.
Speaker B:Yeah, and the newsletter also mentions Perplexity's browser Comet and these new computer use agents.
Speaker A:Oh, right, the ones replacing what agencies used to charge four figure retainers for.
Speaker B:Exactly.
Speaker B:But I think a lot of people still don't fully grasp what a computer use agent actually does.
Speaker B:Mechanically.
Speaker A:So a standard LLM lives in a chat box.
Speaker A:You type, it types back.
Speaker B:But a computer use agent breaks out of the chat box mechanically.
Speaker B:It is giving control of your cursor and keyboard.
Speaker A:It can literally see your screen by parsing the visual data and the underlying code of whatever software you have open.
Speaker B:Right.
Speaker B:So if you tell it to cross reference that 57 page compliance handbook with a live government database, it acts like a human.
Speaker A:It will physically open a browser tab, navigate to the legal site and scroll through the text.
Speaker B:It highlights the relevant laws, switches back to the word document and makes the edits.
Speaker B:And the author notes that configuring these agents used to take a full week of specialized engineering.
Speaker A:Now it takes an afternoon.
Speaker B:Just an afternoon.
Speaker A:I had a moment exactly like this just a few weeks ago.
Speaker A:I had to do this massive spreadsheet analysis combining three different data sets, finding discrepancies, formatting the report.
Speaker B:Oh, the classic weekend killer project.
Speaker A:Yeah, totally.
Speaker A:It was the kind of thing that used to eat my entire weekend.
Speaker A:I fed it into an agent and it was completely finished, beautifully formatted in about three minutes.
Speaker B:And how did that feel?
Speaker A:Well, my immediate gut reaction wasn't joy, it was panic.
Speaker A:I felt this overwhelming urge to wait at least two days before sending it to the client, just to make it seem like it was hard work.
Speaker B:Yeah, you experienced exactly what the author calls the uncomfortable bit.
Speaker A:The uncomfortable bit.
Speaker B:When 72 hours of work collapses into one hour, the problem is no longer the tool.
Speaker B:You have a massive pricing problem because.
Speaker A:If you are billing hourly, you just became 95% less profitable by being better at your job.
Speaker B:This raises an important question, and the core argument in the newsletter is super blunt about it.
Speaker B:We pay for results, not ours.
Speaker A: still charging by the hour in: Speaker A:Sure, result over hours sounds great on paper, but if you are the listener right now, sitting across from a client, how do you actually justify handing them a massive invoice when they know it only took you an hour?
Speaker A:It's tough.
Speaker B:Clients inherently want to pay for sweat equity.
Speaker B:There is a deep psychological comfort for a client in knowing you bled a little bit for their project.
Speaker A:Oh, absolutely.
Speaker A:If the chef prepares the meal in three seconds, the diner feels cheated even if it tastes amazing.
Speaker B:Right?
Speaker A:That psychology is so deeply ingrained in us.
Speaker A:For a century, the billable hour has been our proxy for value.
Speaker A:We equated time spent with value created.
Speaker B:And now that's just gone.
Speaker A:The author challenges the reader with a direct prompt, which I love.
Speaker A:What in your work has quietly collapsed from days to hours this month and what are you charging for it?
Speaker B:They advised a reader to sit with that discomfort because discomfort is information.
Speaker A:Yeah, discomfort is your intuition telling you that the ground underneath your business model has fundamentally shifted, even if your conscious.
Speaker B:Brain hasn't figured out the new math yet.
Speaker A:The cognitive dissonance comes from realizing your value proposition is no longer tied to your endurance or your calendar.
Speaker B:Your value is now entirely in your domain.
Speaker B:Expertise.
Speaker B:Knowing what that 57 page compliance manual needs to achieve legally, and possessing the ability to steer the metacognitive agent to that exact outcome.
Speaker A:But if the execution takes no time and the tool does all the heavy lifting.
Speaker A:The entire apprenticeship model just breaks down.
Speaker B:It does.
Speaker B:You can't even teach a junior employee or a student how to do the work if the workflow itself vanishes into an automated background process.
Speaker A:The newsletter author hits that wall head on.
Speaker A:They talk about the death of static teaching.
Speaker B:They admit to feeling deeply uncomfortable selling prepackaged courses over the past few months, knowing the methods were being rewritten underneath them as they spoke.
Speaker A:Here's where it gets really interesting.
Speaker A:Selling a prepackaged step by step AI course right now is like selling a printed bound paper map for a city where the physical streets literally rearrange themselves every Tuesday.
Speaker B:That is the perfect analogy.
Speaker B:By the time the ink dries, the landmarks have moved.
Speaker A:Exactly.
Speaker A:That is the reality of the current iteration cycle.
Speaker A:By the time you research the modules, record the videos and launch the digital course, the specific prompting technique you are.
Speaker B:Teaching is obsolete because the agent now just does it natively.
Speaker B:The author's solution is a complete pivot.
Speaker B:They say to stop trying to prepackage what is moving and shift to building alongside people in real time.
Speaker A:Moving from a teacher of past methods to a co builder of present solutions.
Speaker B:Which makes sense for anyone in education.
Speaker A:Or consulting, but it makes me look very closely at the specific audience this newsletter is targeting.
Speaker A:NLP practitioners, right?
Speaker B:Master practitioners, trainers, People who study the structure of language and human change.
Speaker A:Why is the author sounding the alarm for them?
Speaker B:Specifically because this is the cerebral core of the document.
Speaker B:The author is talking about actively shaping the foundational psychology of artificial intelligence.
Speaker A:The newsletter asserts that the underlying patterns AI uses to reason are shifting right now.
Speaker B:The warning is stark.
Speaker B:What gets reinforced now becomes the default later.
Speaker A:Let's translate that for the listener.
Speaker A:Because the leap from Claude catches its own faults to NLP practitioners are shaping AI psychology requires us to understand how these models actually learn.
Speaker B:Yeah, we need to look at RLHF reinforcement.
Speaker B:Learning from human feedback, right?
Speaker A:When an AI model is initially trained on the entire Internet, it is basically an alien brain.
Speaker A:It doesn't know how to talk to us at all.
Speaker B:To make it conversational, companies use RLHF.
Speaker B:They have thousands of human gig workers read the AI's responses and literally click thumbs up or thumbs down.
Speaker A:They reward the AI for being helpful, polite and harmless.
Speaker B:So the AI learns to adopt this very generic, slightly overly enthusiastic customer service.
Speaker A:Voice because that is what the average raters gave a thumbs up to.
Speaker B:Which makes sense.
Speaker B:The average rater is just looking for a correct answer.
Speaker A:But an NLP practitioner is doesn't just look at the what of A sentence.
Speaker A:They look at the how they study the architecture of influence.
Speaker B:Like how do you pace a conversation to facilitate actual human understanding?
Speaker A:How do you structure an argument to bypass cognitive bias?
Speaker A:How do you use language to alter a person's state of mind?
Speaker B:They aren't just giving a thumbs up, they are structuring prompts and feedback loops that fundamentally alter the AI's shape of language and decision making logic.
Speaker A:But wait, hasn't the concrete already dried?
Speaker A:Anthropic and OpenAI have been doing RLHF for years.
Speaker A:Isn't this narrow window the author is talking about already closed?
Speaker B:Well, the concrete of pure capability is dried.
Speaker B:The models know how to code, write and process data, sure.
Speaker B:But the concrete of human centric decision making, how the model chooses to present that data to a human being is still wet.
Speaker A:Because AI can process a million data points, but if it just dumps them on you, it's useless.
Speaker B:Exactly.
Speaker B:The author argues that redirecting these massive systems later will be exponentially harder.
Speaker A:AI will not become what we declare, the newsletter states.
Speaker A:It will become what we reinforce.
Speaker B:Right now, somebody is reinforcing something.
Speaker B:The only question is whether it is.
Speaker A:You that is heavy.
Speaker A:You can't just tweet that AI should be ethical or empathetic.
Speaker A:You have to mechanically build the structures that make it so.
Speaker A:Right, but how do you actually do that mechanically?
Speaker A:How are they extracting a human's underlying decision making matrix and putting it into a large language model?
Speaker A:It's tricky because right now, I mean, most people just point a custom GPT at their existing blog posts, tell it to sound like me, and call it a day.
Speaker B:The author draws a massive distinction here.
Speaker B:In the PS section of the newsletter, they pitch a live session called Building your voice and AI through modeling, and.
Speaker A:They contrast an Internet sweep with true NLP modeling.
Speaker B:An Internet sweep being the custom GPT scenario you just mentioned, right?
Speaker A:An Internet sweep scrapes your past emails, your social media, your published articles.
Speaker A:It analyzes the frequency of your vocabulary and basically creates a remix.
Speaker B:It might use your favorite catchphrases, but it is entirely superficial.
Speaker B:It is just an echo of your past outputs.
Speaker A:True NLP modeling works completely differently though.
Speaker B:In NLP modeling is the process of mapping a person's internal representations.
Speaker B:It asks, what is your internal dialogue before you write a sentence?
Speaker A:How do you weigh risks?
Speaker A:What beliefs must you hold to make the decisions you actually make?
Speaker B:So instead of just feeding the AI the final blog post, you are feeding it the cognitive engine that generated the blog post.
Speaker A:You are encoding the metadata of your own thought process into the system prompt or the fine tuning weights.
Speaker B:It is the difference between an AI that knows what you said yesterday and an AI that knows exactly how you would approach a brand new unprecedented problem tomorrow.
Speaker A:The author is using this live demo to walk people through capturing their identity at depth and encoding it into an AI agent.
Speaker B:And there's this brilliant little meta layer at the very bottom of the document that proves the author actually practices what they preach.
Speaker A:I noticed that they actually list the subject line options they considered for the newsletter itself, showing their own internal NLP matrix at work.
Speaker B:Yeah, they list four options.
Speaker B:Option one three days, one hour.
Speaker B:You do the math.
Speaker B:The author notes this is the sharpest business LED option.
Speaker A:And option four, the window is narrower than you think, which they categorize as urgency LED lands on the close.
Speaker B:They are actively demonstrating the exact pattern awareness they want to teach the AI.
Speaker A:If we connect this to the bigger picture, it ties the entire philosophy of the document together perfectly.
Speaker B:The author identified the time collapse, highlighted the crisis of the billable hour and diagnose the failure of static teaching.
Speaker A:And their solution.
Speaker A:This new membership where they walk through building a deep identity level AI voice agent alongside you, solves the pricing problem by creating a high value non hourly asset.
Speaker B:It solves the philosophical problem too by ensuring that highly structured human first intelligence is actively shaping the AI.
Speaker A:It is an incredibly elegant synthesis.
Speaker A:We started with the shocking reality of a 57 page legal document being rewritten in an hour, but by an AI utilizing actual metacognition.
Speaker B:We explored how that sheer speed completely breaks the psychology of the billable hour, forcing a very uncomfortable but necessary pivot toward pricing based on results and domain expertise.
Speaker A:We looked at why selling static prepackaged knowledge is a losing game when the technological landscape shifts daily.
Speaker B:We broke down the urgent narrowing window for NLP practitioners to hijack the RLHF.
Speaker A:Process, right to embed deep structures of human influence into AI before default behaviors harden permanently.
Speaker B:And finally we saw the mechanical difference between a superficial Internet sweep and deep NLP modeling that captures true human identity.
Speaker A:So look at your own calendar for this week.
Speaker A:Where is your discomfort telling you that your pricing or your process is suddenly obsolete?
Speaker B:Where are you still trying to sell a printed map of a shifting city?
Speaker A:And as you think about that, there is one final lingering thought from all this talk about true NLP modeling.
Speaker B:The goal is to capture who we actually are beneath the polished surface of our professional content.
Speaker B:To encode our true decision making matrix into an AI agent.
Speaker A:But human psychology is incredibly messy.
Speaker B:Yeah, it is if this modeling truly works, what happens when we inevitably encode our own unexamined flaws into these agents?
Speaker A:That's a terrifying thought.
Speaker A:If you build an AI voice agent that captures your true deep identity, will your new digital clone also inherit your imposter syndrome?
Speaker B:Will replicate your hesitations, your blind spots, your anxieties?
Speaker A:And if thousands of practitioners do this successfully, are we saving the future of AI?
Speaker A:Or are we just permanently hard coding humanity's deepest neuroses into the global infrastructure of tomorrow?
Speaker B:That is a staggering reality to consider as these tools get closer to mirroring our actual minds.
Speaker A:Keep questioning the default.
Speaker A:Keep paying attention to that discomfort, and we will catch you on the next Deep Dive.
