I have noticed a phenomenon, should I say ‘a pandemic’, for some months now. 

Whenever I have to read blogs on software engineering tutorials or topics, it feels like what I am currently reading is almost the same as what I read earlier.

That was when it clocked, “wait a minute, how come everyone sounds the same?”

Well, the answer isn’t far-fetched. It is most likely because many teams now run their content with AI agents. 

The purpose of this short blog is not to argue whether or not running your content engine with LLMs is good or bad; that is merely academic.

Instead, I want to highlight a few factors that were not spelt out loud, as well as how modern authentic devtool companies can stand out in the age of slop.

But let’s build our case…

Many Devtool Companies Dish AI Content Because It’s Cheap

At the end of the day, everything is based on the law of demand and supply. Companies pay writers because they can help them create product-led content.

Now, tides have changed. LLMs can also spin up articles, LinkedIn posts, threads, whitepapers, and whatever you could have paid a writer for.

Interestingly, there are not only different LLMs like Claude and ChatGPT, there are also various models like Fable and Opus. 

So there are quite a couple of different options on the AI content side of things.

And this is the part we have to talk about: creating content with LLM is very cheap. Everyone can afford it, even solo founders who have less than $500 MRR on their dashboards.

For this reason, many companies don’t see the reason to pay an agency or freelance writer to do what an LLM can equally do.

Indeed, putting a content agency on retainer generally never costs less than $5k. Whereas, with a $12 subscription, anyone can create quite an unlimited amount of content with ChatGPT.

Many teams even go as far as firing their SEO and content teams. Whereas, a prominent AI company, such as Anthropic was speedily hiring copywriters some months back.

Substituting Affordability for Depth, Taste, and Conversion

When anyone says, “using AI is definitely cheaper, better, and faster than hiring an agency.” It raises a question of what we really think is great content.

If your judgement of great content is anything that is lengthy enough till 1,500 words, then we know there is a problem.

From a wider perspective, the main goal of creating technical content is to communicate, propagate, and convert. The goal is visibility, education, and conversion, content creation is only a means to that end.

However, the link is broken when creating technical content with AI. Why? 

A good number of AI content, even those created with the best models, are objectively bland and obviously lack depth.

After reading a whole article, you just discover that there is quite no contribution to knowledge, and even though it is verbose, next to nothing is really communicated.

The problem is, once a reader sets out to read an article, and they end up not gaining what they thought they would, then the value chain is broken ab initio.

Yes, AI content can be very cheap enough, but they don’t fly. 

In an alternate universe, this is the same reason you don’t see anyone with a vibe-coded startup ever claim they hit 1k paid users.

Understandably, perhaps that’s the reason startups like HoneyBadger came up with their human-centered content policy.

Models Are Trained on Same Data; Hence, Same Response

LLMs are built on models, and models on data.

Here is a quick explanation: there are 20 available research papers on summary judgement in the United States, and a model is trained with them.

This model will only know as far as the 20 papers are concerned. It will also believe erroneous procedures or facts in these papers it is trained with.

That said, you cannot prompt your way for this model to give you an accurate summary judgement procedure in a jurisdiction like The Netherlands. Why? It was only trained with content from the States.

I’ve heard a couple of people say, “Ohhh, all you have to know is prompt engineering. If you prompt it well enough, you’ll get more from it.”

From a technical standpoint, this is a flat lie. It doesn’t work that way. A model doesn’t know more than what it is fed, at least for now.

What “prompt engineering” does is to provide more specific contexts, so the model can know more specific answer from its dataset to give as the answer.

Do you see the flow of thoughts now?

Now, that explains why everything has been reading the same. No, it won’t be the same word for word simplicita, but these things will be the same:

• the overall message
• a good number of expressions
• structure

Technical Experts Remain the Best Bet for Technical Content

Content in the developer space is even more niched and with its peculiarities. For example, developers are go-getters. 

If an article says it will help connect a smart contract to a backend database, by the end of reading the article, the reader must be able to replicate the same thing, and also understand the nitty-gritty from the beginning to the end.

For this reason, there is a level of details required in properly drafting technical content.

And this is why practicing developers who are product advocates are the best to write about these things, and I dare say, “not an AI.”

Again, AI is very fantastic for search; it will be pretense to disagree. But when it comes to depth and niche-level head-to-head, it’s not the best companion.

Technical writers know the tools that have shut down, the practice to encourage, and fluff to cut off before an article goes live.

For example, in Web3 engineering, we have stopped using .env for security reasons. Instead, we now encode secrets in our CLI.

Conversely, a good number of LLMs still recommend this old practice that can get your wallet drained in a second.

Have you wondered why your fave companies still hire technical writers and SEOs?

As someone who’s always in support of innovation, I’m well in support of AI, and I want to contribute to training models. Fun fact, I’ve been a fan of AI since the Internet of Things (IoT) days.

Notwithstanding, I’m mature enough to realize that the proportion of AI now, as proclaimed by many, is honestly overblown.

And that is fine, it happens with any new innovation; do you remember NFTs? What I’m more after is the fact, and treating things exactly as they are, and not more.

If AI content generation were as powerful as portrayed, how come big companies still hire for these same roles?

MongoDB, Bitwarden, n8n, and many others are on a hiring spree.

Bringing Back Taste and Confidence

One of the reasons startups are hiring more technical writers in their content strategy is because they want someone who’s been there and done that to create their content.

This leads us to taste; a practising software engineer will know for sure what other engineers face, and how to tailor content in a way that fixes their problems, while still preaching the product in question.

In the age where every blog reads the same, those with taste will win. How do you know when a piece of technical content is created with taste?

Simple answers, when you notice a good amount of craftsmanship was deployed to create it. Really, it will all be noticeable from the first glance.

You’ll enjoy the organic flow of thoughts, the apt examples, and tight code blocks to explain things in-depth. In some other blogs of this nature, you’ll even see diagrams, and all the code blocks packed into a GitHub gist.

There is another factor that showcases taste: confidence.

Perhaps one of the things that made you read this blog this far was because of this very factor: you noticed I didn’t mince words with everything I was saying, and the confidence was glaring.

Yes, readers love that.

But confidence doesn’t come from models. It comes from experience.

As a software engineer who has built and shipped things to production, you will always be confident to show it to others.

In turn, the reader will also feel like they are in the safe hands of someone who really knows what he is saying.

To have a better feel of what I’m saying, go read some old articles on the DigitalOcean blog and see for yourself.

Not Losing Sight of the End-game: Revenue

In every startup, each department must be tied to revenue as their ultimate yardstick; that is simply how it works.

In the same way, content marketing departments were created to cause visibility and influence conversion. 

All the technical tutorials and guides are to the end that customers get to know their product, and somehow along the line become paid users.

In another blog, I’ll explain the siamese nature of content marketing and sales teams, and how they pass the baton to each other.

Back to the discussion, AI generated content often lacks contribution to knowledge, taste, and confidence.

As a result, readers don’t trust it. And since they are unimpressed, they wouldn’t bother to check out the product of such startups.

This is how you lose leads.

In turn, this is why devtool startups in the game know they have to bring something more concrete to the table, so they won’t lose the money on the table.

Concluding Thoughts

This is the summary of everything canvassed in this short blog:

The progress some frontier labs, such as Anthropic and OpenAI, have made regarding the training of models has been nothing but amazing. Massive respect to them.

Notwithstanding, AI content generation doesn’t objectively add sufficient value to readers. It lacks adequate contribution to knowledge and taste.

A startup founder who wants their dev blog to be outstanding? You have to invest in hiring a technical content agency like Blockchain Alpha, or perhaps an in-house technical content engineer.

Either way, you need a solid content department on your team.