Where ChatGPT Actually Sits in 2026: Hype, Reality, and What It Replaced

This article was originally written in 2023 about ChatGPT's hype cycle. I updated it in May 2026 because the question worth answering changed: not 'is it overhyped' but 'what did it actually replace, what does it cost, and where does it fit now.' The original argument is at the bottom for context.

The 30-second 2026 update

The hype cycle around ChatGPT peaked in early 2023, dipped through the 'is this just a toy' phase in 2024, and has settled into something more interesting in 2026: a tool a billion people use weekly, that has quietly absorbed several adjacent industries, and that has become a foundational piece of how software is now built.

Was it overhyped in 2023? Yes — in the short term. The promised tsunami of layoffs and AGI in 18 months didn't happen on that timeline.

Was it underhyped in 2023? Also yes — in the long term. What we got over three years made the original predictions look conservative in some ways (coding, design, content) and absurd in others (autonomous business operations, full creative replacement).

I run an AI venture studio at Shape. Here's what the picture actually looks like in 2026.

What ChatGPT actually replaced (and didn't)

Replaced almost entirely

• First-draft writing. Marketing copy, internal memos, emails, basic blog posts. The cost of a 'good enough' first draft went to zero.
• Translation between major languages. Specialized tools still beat it in narrow domains, but for the long tail of business communication, ChatGPT replaced an entire freelance category.
• Coding boilerplate. CRUD endpoints, configuration files, test scaffolds, type definitions. The 80% of code nobody wanted to write is now written by Claude or ChatGPT instead.
• Customer-support tier 1. The 'reset password / track my order / where's my refund' tier is mostly automated. The complaints, the edge cases, and the angry users still need humans.

Partially replaced

• Research and analysis. ChatGPT is great at summarizing 100 pages into 5. It's still not great at noticing what's missing from those 100 pages. Senior analysts still beat it.
• Design ideation. The first 20 ideas are now ChatGPT-generated. The decision about which one is right for the customer is still human.
• Tutoring and explanation. Genuinely transformative for students, but only the ones who already know how to ask good questions. The gap between 'good user' and 'mediocre user' is enormous.

Did not replace

• Original strategy. ChatGPT does not know what your business should bet on next quarter. It can talk about what others have done. That's not the same thing.
• Trust-based work. Therapy, complex sales, leadership decisions. Anything where the human relationship is the value.
• Creative work where taste matters. The ChatGPT-generated essay or song or image is increasingly indistinguishable from average human work. It is reliably worse than top-tier human work because it has no taste, only consensus.

Where the 2023 hype was right

Looking back, three predictions from the peak hype phase ended up roughly correct.

'It will eat the search box'

This took longer than predicted but it happened. By 2026, a meaningful share of search traffic that used to go to Google goes to ChatGPT, Claude, Perplexity, or Gemini directly. Publishers are still figuring out what their business model looks like in that world.

'Software development will be transformed'

Yes, and faster than the median engineer expected. Agentic software development is the new default. Teams that adopted it early ship 4–6x more code per engineer than teams that didn't.

'It will commoditize content production'

True. The cost of a 'fine' blog post, summary, ad, or email went to zero. Premium content (deep reporting, original research, real expertise) became more valuable, not less. The middle disappeared.

Where the 2023 hype was wrong

'AGI in 18 months'

No. We are demonstrably not there in May 2026. ChatGPT is extraordinarily capable at language tasks. It is still bad at reasoning that requires sustained planning, novel logical inference, or honest uncertainty about its own knowledge.

'It will replace 50% of jobs by 2025'

It replaced parts of jobs, not jobs. The mistake was treating 'a task that AI can do' as 'a job AI can do.' Most jobs are bundles of tasks, only some of which are LLM-amenable.

'Hallucinations will be solved by next year'

Reduced — not solved. Modern models are dramatically better at staying grounded in source material, but confidently wrong outputs still happen and the only fix is verification, not prayer.

What ChatGPT looks like in 2026

Concretely:

• Default model: GPT-5 generation by default; OpenAI's stack now includes routing across smaller and larger models invisibly.
• Multimodal by default: images, audio, video, code — all in the same conversation. The 'text-in, text-out' framing of 2023 is gone.
• Agents: ChatGPT can take actions on the web on your behalf. Booking, research, data entry, coding tasks. Still rough on long-horizon agentic work but improving fast.
• Memory: persistent memory across sessions is now standard. Privacy implications are still being negotiated.
• Pricing: $20/month for Plus stayed stable. Higher tiers ($200+) for power users and teams. API costs continued to drop ~3x per year, making AI-native products economically viable that weren't in 2023.

What this means for builders in 2026

If you're shipping software now, three things matter more than they did in 2023.

1. ChatGPT is your competitor's commodity, so don't compete with it

Wrappers around ChatGPT have nearly zero defensibility. Build products where the AI is one ingredient and the workflow, the data, the integrations, or the eval framework are the moat.

2. Use it to ship, not to think

ChatGPT and Claude are extraordinary execution multipliers. They are not strategy generators. The thing they do worst is decide what you should bet on. Don't outsource that.

3. Build for what's next, not what's now

Models continue to get cheaper, faster, and more capable. Whatever expensive product you're building today will be a feature in someone else's app in 18 months. Build the moat in customer relationships, data, or integration depth — not in clever prompting.

The honest verdict

Was ChatGPT overhyped in 2023? In the way most overhyped things are: people overestimated the next 18 months and underestimated the next three years. The result is that almost every prediction made at the peak was wrong, but in different directions than the critics expected.

In 2026 it's not overhyped or underhyped — it's installed. A foundational piece of how knowledge work happens, the way email and search engines became foundational in earlier eras. The interesting questions aren't about ChatGPT itself anymore. They're about what you build on top of it.

If you're building an AI-native product and want to talk through what's a moat and what's a wrapper, book a 30-minute call.

Below: the original 2023 article, kept for posterity.

Original 2023 take: Was ChatGPT overhyped?

When ChatGPT launched in late 2022, the response was somewhere between 'this changes everything' and 'this is the end of work as we know it.' Most articles from that period either celebrated it as the dawn of AGI or warned about civilizational risk.

My 2023 view was simpler: it was a remarkable language model, surprisingly useful for first drafts, dangerously confident when wrong, and the hype was overshooting the practical reality at the time. It was a tool, not a revolution — and the people who would benefit most were those who treated it as one.

Three years later, that view holds in spirit but understates the long-term impact. Tools that are genuinely useful for first drafts and good enough for tier-1 customer support do, in fact, change a lot of jobs over time. Just slower than the hype cycle predicted and faster than the skeptics expected.

Written by Marko Balažic, founder of Shape — an AI venture studio building AI-native software for founders. If you're working on something interesting, reach out.