PRODUCTHEAD: The value of the human in the loop

products out #

every PRODUCTHEAD edition is online for you to refer back to

---

tl;dr

Semantic ablation is when language outliers and quirks are averaged away by genAI

GenAI exposes the thresholds and grey areas in your service’s decision-making

Being able to build fast necessitates more human selectivity, not less

---

hello

I’m beginning to suspect that the kind of corporate leaders who wave away swathes of employees in favour of generative AI are the kind that value mediocrity.

Perhaps they’ve never experienced anything else. Maybe they’re perfectly content with feeding a hodge-podge of requirements into a metaphorical meat grinder and don’t care as long as a vaguely sausage-shaped product emerges at the end of the process. And when the process is thoughtless and there’s no desire to change it, what difference does it make whether it’s a product person or a LLM-powered bot cranking the handle?

Not all product people are in roles where they’re expected to exercise their emotional intelligence and creativity. Their bosses are more interested in throughput and efficiency rather than the quality of the actual work and the real-world effect it has – they want their employees to act like unthinking machines. If you work somewhere like this, you’re never going to be asked to work any differently. And yes, this means there is a real risk of your replacement by generative AI.

A couple of chats I’ve had this week have explored the value a product person brings to an organisation. Perhaps generative AI highlights the value of having a product manager as the ‘human in the loop’, precisely because we can do what genAI cannot:

our capacity for empathy and true human connection (not simulated sycophancy);

to ask the insightful questions that change our perspective and understanding;

for exercising informed judgement in an attempt to balance the inevitable nuance and trade-offs inherent in working life;

for asking ‘should we?’ as often as ‘can we?’;

for making and, importantly, learning from mistakes;

for creativity in the face of constraints.

All of these set us apart from statistically amalgamated responses to prompts.

It’s quicker, easier and more tempting than ever before to ignore your users and build something mediocre that nobody needs. Your challenge is therefore to elevate the work without being asked to, to go beyond the brief and re-introduce the value that comes from understanding the people in the system and, with that understanding, to build what they actually need.

Even if your extra efforts remain unappreciated, you’ll have the satisfaction that you’re doing your best to honour your craft. And with any luck you’ll take your experience and move to a different organisation that values the human in the loop.

For you this week #

In an opinion piece for The Register, Claudio Nastruzzi describes why AI-generated content seems anodyne and ‘off’. He calls it ‘semantic ablation’, and to paraphrase slightly, it’s the result of a LLM’s propensity to average out nuanced and edge-case content in favour of the generic. ‘If “hallucination” describes the AI seeing what isn’t there, semantic ablation describes the AI destroying what is.’

Tom Wynne-Morgan considers how genAI is going to change customer service. Decisions used to rely on hard-coded rules: ‘You don’t meet the criteria, your application is rejected’, or as it became colloquially known in the UK, ‘Computer says no’. Now that the computer is going to return probabilistic assessments (‘this payment is potentially fraudulent with 63% confidence’), a human needs to decide where the threshold for the decision is.

Next a couple of posts looking at how AI is exposing different bottlenecks in the system of bringing products to market. When coding can go fast, Rich Mironov observes that we have to be even more selective of what we choose to build, rather than succumbing to the temptation to ship the backlog.

Similarly, Cleo Lant notes that users have only a finite amount of attention to absorb the new features you’re marketing. Trumpet everything you’re now shipping at light speed and even your most engaged users will be bamboozled by the flood of new things. Be selective.

Speak to you soon,

Jock

---

---

what to think about this week

Why AI writing is so generic, boring, and dangerous: Semantic ablation

Just as the community adopted the term “hallucination” to describe additive errors, we must now codify its far more insidious counterpart: semantic ablation.

Semantic ablation is the algorithmic erosion of high-entropy information. Technically, it is not a “bug” but a structural byproduct of greedy decoding and RLHF (reinforcement learning from human feedback).

A civilizational race to the middle

[Claudio Nastruzzi / The Register]

Computer Says Maybe

For years, customer service had a kind of structural clarity. The system held the rules. The agent applied them. The outcome was binary.

Generative AI shifts that arrangement. Instead of hard-coded rules, we increasingly see probabilistic assessments.

We can’t hide behind the machine’s decisions any more

[Tom Wynne-Morgan / Medium]

---

---

Bottlenecks, AI, and Where Product Adds Value

Let’s imagine that AI development tools actually improve overall software development productivity by 15x. Where are the next bottlenecks, and how would that change what product managers do?

We get paid to sort wheat from chaff

[Rich Mironov / Product Bytes]

The hidden danger of shipping fast

Is it possible to ship too much – or too fast?

Yes. Probably. Unfortunately.

A handful of people with good judgment and a lot of tokens can now do what used to take a full product org. As a result, software powered by LLMs is cheaper to build and scaling faster than at any point in history.

What to do when product velocity breaks the speed of adoption

[Cleo Lant / Posthog]

recent posts

Canary in the mine: AAA game developers are unionising

Product management has had its own fair share of problems over the last few years. Nevertheless, there are early warning signs from AAA game studios that there may be another storm brewing in tech for us to weather.

Union-busting just isn’t a good look

[I Manage Products]

Startup to Scale-up Club Q&A – 13th Jan 2026

I joined Anton Kooll again, along with co-panellists Maarten Ectors, Mario Tomic and Eugenio Galioto for Startup to Scale-up Club. We covered topics including:

• recommendations for automated infrastructure monitoring;
• safe presentation of medical data in Femtech apps;
• trade-offs of cloud versus local AI deployment for agricultural technology;
• the risk of patronising, gender-based marketing in Fintech; and
• the value of building a trustworthy community over superficial personalisation.

“Do’s” and “don’ts” for startup founders

[I Manage Products]

Are developers vibe coding themselves out of a job?

And is the increasing reliance by junior developers on AI coding assistants storing up a generational skills shortage for the future – ‘professional debt’, if you will?

So simple, anyone could do it. Wait – don’t fire me

[I Manage Products]

PRODUCTHEAD is a newsletter for product people of all varieties, and is lovingly crafted from a candy floss sundae.