The specialists: images, search, code and open models

The big three cover almost everything you'll run into. Almost — and this lesson is about the remainder.

Specialists are needed less often than the internet suggests, but when they are, an all-rounder won't stand in for them. It's worth knowing their faces in advance, so that at the right moment you're not googling "what draws a decent picture".

Three stories instead of a catalogue

A list of names is forgotten in a day; stories aren't. So let's meet them through three moments when specialists made the news.

March 2023: the Pope in a puffer jacket. A photo of the pontiff in a white designer puffer went round the internet — stylish, expensive, utterly unlike him. Millions believed it was real, including people who normally believe nothing. The picture was generated in Midjourney by a construction worker in Chicago who had an ordinary evening and time on his hands. That viral fake demonstrated the state of artistic AI better than any advert: the light, the fabric, the creases all looked like a real photograph. That's exactly what designers pay Midjourney separately for.

January 2025: the DeepSeek shock. A Chinese lab published an open model, R1, at flagship level and said it had trained the thing for a fraction of the usual cost. The market reacted instantly: on 27 January 2025 Nvidia — the company whose graphics cards train most of the world's AI — saw its shares fell 17% in a day, costing the company $589 billion — the largest one-day loss of value in US market history. The moral for you is simpler than the market reports: open models aren't hobbyist toys, they're a force that gets taken seriously.

Perplexity and its footnotes. A 2022 startup bet everything on one thing: every statement comes with a link to its source. Not "sounds plausible" but "here's where this comes from, go and check". Analysts, students and anyone who has to answer for what they write loved it for exactly that.

Who's who among the specialists

Now in order, with one caveat: nearly all of them have a free way in, and you don't need to start any subscriptions today.

Images — Midjourney. The benchmark for artistic generation: style, light, composition. Designers and marketers keep it as a separate subscription. To begin with, the generators inside ChatGPT and Gemini will do: for a social post or a slide they're usually plenty. You want Midjourney when the image has to look like a film poster.

Search with sources — Perplexity. A hybrid of an AI and a search engine. Indispensable when what matters isn't a nice write-up but where the claim came from. The free tier is generous and enough for fact-checking.

Open models — DeepSeek and relatives. An open model means its weights — the result of training — are published, so you can download and run it yourself. The training code and data usually stay closed. Two consequences follow: it's free or very cheap, and your documents go nowhere. It's the choice of companies with sensitive data and of people for whom privacy isn't an empty word.

Code — Cursor, Claude Code, Copilot. Tools where the AI writes code inside the project itself rather than in a chat window. You'll meet them if you go into vibe coding — building a site or app from a text description — and we have a separate course on precisely that.

When to call a specialist

The rule is simple and it saves both money and time: start with an all-rounder, call a specialist when you get stuck.

Stuck means three specific situations. You need an artistic image rather than a filler illustration — Midjourney. You need sources, because you'll have to answer for the text — Perplexity. You need the data never to leave your own server — open models.

In every other case an all-rounder will cope, and there's no reason to open a second account.

Where people usually start, by role

If you want to picture your own combo straight away, look at what tends to form for people with similar work.

• Content and social: an all-rounder for text plus Midjourney for visuals, once images become part of the job rather than decoration.
• Analysis and study: an all-rounder for drafts plus Perplexity when you need facts you won't be embarrassed to show a supervisor or a client.
• Building things with no experience: an all-rounder plus vibe-coding tools — that's the combo people with no training use to put working websites together.

You can try them free, today

The good news: getting acquainted with specialists needs no new subscription at all.

For artistic images, look at the generators inside ChatGPT and Gemini — for posts and slides they're more than enough. For search with sources, try Perplexity's free tier; the big three have web-search modes too. And for music, listen to what Suno does: you describe a mood and a couple of lines of lyrics and get back a finished song with vocals. That one impresses even the people who shrugged at everything else.

What we have so far, and what's ahead

Half the map is done: the big three, their characters, the money, and the specialists round the edges. The second half is more practical — a table for picking a model per job, AI on your phone, what you can and can't hand it, and where it lies.

Do this now

Pick one five-minute experiment to find your own border between "the all-rounder is fine" and "time to call a specialist".

Either ask an all-rounder to generate an image for your most recent post or document, and judge honestly whether it's usable. Or ask Perplexity a question from your own work and follow a couple of the sources in the answer — see whether they really say what it claims.

A border you found on your own task is worth more than any review: reviews show the best case, and what you need is yours.