AI in 2025 + the era of freelancing and consulting

One clear trend standing out since 2024 is the empowerment of highly skilled individuals mastering LLMs and the corresponding wrapper apps.

First, some light 101 recap on practical AI in 2025

AI is not ChatGPT

Still fascinated that I keep hearing that times and again during interviews, but I’ve seen over 200 CVs lately claiming AI literacy or “using AI” that usually boils down to “I ask ChatGPT every other day when I get stuck”.

This isn’t AI literacy.

AI and LLMs 101

First, the foundation of AI was started in the 1950s, with earlier actions as early as the 20s and 30s with Alan Turing. Different branches have been deployed in production over the past 50 years, including machine learning algorithms, natural language processing, computer vision, robotics, and important sub-branches like neural networks and deep learning.

LLMs are designed for NLP tasks and also a branch of deep learning.

Second, there’s a handful of powerful LLMs in use by OpenAI (GPT-4o, o1, o3 etc), Anthropic (Claude’s models), Google (Gemini’s models), Meta (LLama), DeepSeek (R1, V3), Mistral, Alibaba, Microsoft, xAI and more.

So basic questions toward a ChatGPT input isn’t constituting AI literacy.

Third, proper discussions with LLMs require the so called “prompt engineering”, or understanding the underlying philosophy of LLMs, how they intercept and perceive data, and how to properly pass context.

Prompt engineering represents the required components to converse with an LLM, including the actual directive, the context, some input data (samples), output format, tone of voice, persona (among others). The more context provided to the engine, the more accurate the answer.

Fourth, LLMs have been expanded into two main categories, agents and wrappers. Agents are autonomous (or semi-autonomous) software solutions that use LLM data to make contextual decisions, adapt based on the circumstances, and provide situational data to solve specific tasks. Agents can also speak to other agents, making it possible to break down complex problems into separate agents responsible for different things.

An example in blog post generation would include an agent for headline generation, one for cluster/topic structure, one for intros, one for actual writing, one for link building, one for image generation, one for research and pulling sources, one for cross-linking internally, and so forth. Each is dependent on the context from the others, and over time, they evolve based on external factors (like GA traffic or SERP fluctuations, and linking to new and old posts as the collection grows).

Some wrappers work as agentic structures while others simply provide prompt engineering templates to better serve specific purposes - by supplementing private data sources for better results.

One of our wrappers is the Offer Consultant - an LLM wrapper fed with the best frameworks for creating irresistible offers, ensuring that simple prompting will still result in powerful offers for users. This trims down the complex layers of searching for the right framework or the criteria for creating great offers, including a subset of great products and services fed to the engine.

So AI proficiency today means understanding the broader set of AI opportunities, and in the context of LLMs, experience using different models, comparing the outputs, understanding the best practices (when to use each), and a good command of a dozen or more wrappers and additional tools to get the job done effectively.

Are AI taking jobs today?

Yes and no.

AI alone does NOT take jobs. LLMs hallucinate a lot, or misread data, and are often fed on public sources (which are inaccurate, poorly written, outdated, or not referencing legitimate sources).

Additionally, great business insights are not based on generic data found online. Similarly to scraping Google for go-to-market strategies, most channels are already oversaturated, and deep understanding of an industry + competitive advantage (relationships, brand moat, proprietary data) are integral to making the entire puzzle fit together.

That said, truly experienced individuals can save a lot of time by scaffolding or generating templates, or supplementing data they know and understand, but need compiled in the right way.

The essence here is that experts know what the output looks like and can attest to the accuracy of the data. Just like software engineers can freely use copilots or tools like Cursor or v0 to scaffold code, non-engineers have no idea what that code means, whether it’s fast, secure, stable, and reliable. And failing to understand can lead to massive security vulnerabilities or exposing API keys that cost thousands of dollars if used by malicious users (to name a few of the common problems with SaaS code today).

But mundane tasks completed by VAs, interns, or juniors are getting easier to replicate fast. This is jeopardizing the work of corporate workers or data entry “specialists” in cases where decent results can be accomplished by tooling today.

This is beneficial to consultants, freelancers, and small experienced teams

As generic and mundane roles can now be automated for the most part, many core tasks can be accomplished much faster when overseen by experienced workers. This means that teams of 10 people can now be squeezed down to 3-4 people, or upskilled so that 10 people get 5X the amount of work done when using LLMs properly.

As a result, individuals can also deliver more. Product professionals, marketers, engineers can take on more roles individually - or work in small teams of 2-3 people getting a lot done together, such as:

1. Building starter SaaS apps way faster with AI

2. Crafting landing pages with AI

3. Generating blog posts at scale with SEObot or programmatic SEO

4. Automating contextual social media content

5. Sending email outbound sequences

6. Crafting images and landing page creatives

And many more.

Even on the consulting side, creating a comprehensive proposal that previously took 20-40 hours of work can now be done in 3-4 hours with proper prompt engineering and templates. Consultants and freelancers can supervise agents and run LLM prompts to get individual tasks done in a couple of hours.

Companies are now looking into more individuals and small teams to outsource “productized” tasks, including email newsletter automation, creating freebies (white papers, ebooks), generating packed infographics (with LLM research), building simplified internal tools, generating UI prototypes for landing pages in a handful of hours.

Instead of hiring a team of 3 FTE people + assigning a manager to the team in house, outsourcing and delegating specific tasks - one-off or recurring - makes a ton of sense today.

Relying on proven expertise + assigning autonomy and responsibility to the experts as external parties is a win-win situation that isn’t tied to payroll and clocking 9-to-5 time. Instead, if the results can be accomplished in a predefined fashion (predictable monthly or task-based cost), both parties are incentivized to maintain the relationship, with the consultants/vendors managing the value equation with updated tooling and refined processes.

The productized model is growing in popularity. Specialize in one or two areas in demand by providing high-quality results in predictable time frame at a fixed cost. Add monthly packages on top. The better you understand the industry and the more optimal your underlying tooling, the more effective and profitable this business will be.