On 9 June 2026, Anthropic released Claude Fable 5, the first model in its new Claude 5 family. The one number worth remembering is from Stripe. The payments company says Fable 5 did a job in a single day that would have taken a whole engineering team more than two months: rewriting part of a codebase with 50 million lines of code.
That is the real story here. Not a smarter chatbot. An AI that can work on a hard software problem for days at a time, mostly on its own, and check its own work as it goes. So the honest question for a Malaysian business owner is simple. If AI can now code for days by itself, do you still need a software house?
Short answer: yes, and arguably more than before. Here is why.
What Claude Fable 5 actually is
Fable 5 is the public, safety-checked version of Anthropic's most powerful model line, codenamed Mythos. (There is a locked-down twin called Mythos 5 kept for vetted security and research partners. Most businesses will never touch it, so we will leave it aside.) Think of Fable 5 as the careful, public release of the strongest engine Anthropic has built.
Two things make it matter for business. First, it is built for long-horizon work, meaning tasks that run for hours or days instead of one quick reply. Second, it is cheaper than the last top model: 10 US dollars per million input tokens and 50 per million output, which Anthropic says is less than half the old price. More capability, lower price. That trend is good news if you buy software, because it keeps pushing the cost of building it down.
Anthropic's own benchmark scores for Fable 5 against other top models. Source: Anthropic
What it can do that you can picture
Forget the benchmarks for a second. Here is what "codes for days" looks like in real life.
Picture your inventory system. The one built years ago, that only your former developer fully understood. A model like Fable 5 can read the whole thing, then rewrite an old payment module overnight and have it tested by morning. That is the Stripe story, shrunk to your size.
Or picture a Friday afternoon. You describe a small internal tool in plain English: a dashboard that pulls sales from your POS and flags slow-moving stock. You set the AI going. It works through the weekend, writes the code, runs its own tests, fixes its own mistakes, and on Monday there is a working first version waiting. Vercel's CTO put it well: the model "understands what builders mean, not just what they type."
It is not only code. Anthropic shows it doing first-pass legal contract review, messy spreadsheet analysis, and research work. For a lean Malaysian SME with no big tech team, that is a genuinely useful junior assistant that never sleeps.
How working with it has changed
Talk to people who use these models every day and the same three shifts come up.
First, treat Claude like a thought partner, not a vending machine. You get more by thinking out loud with it and pushing back on its first answer than by firing off one request and walking away.
Second, give it a clear goal and a way to check its own work. "Build a stock report and make sure every total matches the source data" beats "build a report." The person who can say what "done" looks like, and how to prove it, gets the best results.
Third, be more ambitious. Jobs you would have written off last year as too slow or too costly are now worth a try. A weekend is enough to test an idea that used to need a quarter.
Notice the pattern. All three are human skills: sharper thinking, clearer goals, bigger ambition. The AI got stronger, and what it rewards is the opposite of making you redundant.
This is real, and we are not going to pretend otherwise. At Gotchaa Lab we use these tools every day to build for clients faster. They make a good team much faster. They do not make the team optional. Here is the part the hype skips.
What Claude Fable 5 still cannot do for your business
"Codes for days" cuts both ways. Point it at the wrong goal and it will confidently build the wrong thing for days. Speed without direction is just faster waste.
A few things Fable 5 will not do for you:
- Decide what to build. It will not tell you that your real problem is a broken checkout flow, not a new app. Choosing the right thing to build is judgment, and it comes from understanding your business.
- Handle Malaysian rules. PDPA, where your customer data is allowed to live, the quirks of LHDN e-invoice and MyInvois. The model has no stake in getting these right for you.
- Plug into your messy reality. Your SQL Accounting setup, that 8-year-old system nobody documented, the bank integration with strange edge cases. Connecting to the real world is still slow, careful human work.
- Be accountable. When the system goes down at 2am before a campaign launch, an AI does not pick up the phone. A person who owns the outcome does.
For more on this gap, see our piece on what it means that AI now writes most of Anthropic's own code, and our take on why Claude is overtaking ChatGPT for business use. The lesson is the same: the code is getting automated, the judgment is not.
So what should you do?
Do not panic, and do not buy the hype either. A more capable, cheaper AI is good for you as a buyer. It should make custom software faster to build and easier to afford. The right move is to ask any software partner a plain question: are you using tools like this to pass the savings on to me, and are you still owning the result?
That is the line that matters. A team that uses AI well and stands behind the work is worth more in this new world, not less. A vendor who hides behind "the AI wrote it" when things break is worth less. The technology changed. What you are really paying for did not.
Thinking about how AI fits into your next software project? Let's chat. We will give you an honest take on what to automate and what still needs a human, no sales pitch.
