Loading
Loading
Insights
Not commentary on the industry. The specific things that break, what they cost, and what we would tell you not to build — written from operating our own AI platform in production and from the engagements that came out of it.
The failure modes, in the order we have met them — and almost none of them are the model being wrong.
Read itThe short answer
AI agents rarely fail because the model gives a wrong answer. They fail because the retrieval layer goes stale, because a provider deprecates the model underneath them, because nobody defined what the agent should do when it does not know, because costs scale in a way nobody modelled, and because no one is watching any of it after launch. In production the model is the most reliable component in the system. Everything around it is what breaks.
Four lines of arithmetic that decide whether a project is worth funding — and that most proposals skip.
Read itThe short answer
To cost an AI automation, work out what the current process costs before you look at what the system costs. Multiply volume by minutes per unit by the fully loaded hourly cost, then add rework: error rate times the time to fix. That is your annual baseline. Against it, put the build cost plus twelve months of running cost, where running cost includes model and infrastructure spend, monitoring, and the engineering time to keep it accurate. If the baseline does not comfortably exceed the total, the honest answer is not to build it.
A retrieval system that never says “I don’t know” is not accurate. It is unmeasured.
Read itThe short answer
A retrieval-augmented system should refuse whenever it cannot ground an answer in a retrieved source, and a healthy one refuses more often than teams expect. Refusal is not a failure state — it is the behaviour that makes every other answer trustworthy. If a system never declines, it is not that the corpus covers everything; it is that the system will always produce something, and nobody has measured how often that something is unsupported.
AI answers name one or two companies and send no click. Your analytics cannot see any of it.
Read itThe short answer
When a buyer researches inside an AI engine rather than a search engine, they get a synthesised answer naming one or two companies rather than a page of links — and no click is sent, so the interaction never appears in your analytics. Being cited depends on whether an engine can parse your content, verify your claims and identify you as an entity, which is a structural property of your site rather than a ranking you can bid for. The practical consequence is that a growing share of consideration now happens somewhere you cannot currently measure.
Weeks instead of quarters is real. Which parts compress, which parts refuse to, and where the time goes instead.
Read itThe short answer
Agentic tooling compresses the parts of a software project that were always mechanical: scaffolding, wiring, tests, migrations, documentation and the second and third pass of a design. In practice a working website now takes about a week and a production application two to four, where the same scope took a quarter in 2023. What does not compress is deciding what the thing is for, getting access to the systems it must talk to, and the review of anything the business would be embarrassed to get wrong. Projects that still run long almost never run long because of the code.
Mostly for bad reasons. There is one good one — and either way, here is how to get a real number out of them quickly.
Read itThe short answer
Most software firms hide prices to keep you on a call where the number can be adjusted to what you seem able to pay. That is a bad reason and you should treat it as one. There is a legitimate reason too: the same eight words of brief describe genuinely different amounts of work, so a published figure is either wrong for most buyers or hedged into uselessness. The test that separates the two is simple. Ask for a written figure within one working day, along with what would change it. A firm withholding a price for honest reasons can do that easily. A firm withholding it for leverage will keep trying to book a meeting.
Five things that decide whether a build goes well. None of them is the price.
Read itThe short answer
Before signing a website, app or software quote, get five things in writing: what is explicitly excluded, who will actually build it, what happens when the scope moves, what it costs to run after launch, and what you own at the end. Price is the easiest of the six numbers to compare and the least predictive of the outcome — most projects that go wrong were quoted accurately and scoped badly. A firm that answers all five quickly and in writing has usually done this before; one that keeps routing you back into a meeting is managing you.
Not a chat bubble. A page that answers from your own material, qualifies, books, and knows when to stop.
Read itThe short answer
An AI-enabled website answers a visitor’s question from the business’s own material — prices, policies, stock, past enquiries — then qualifies what they need, books the call or takes the order, and hands over to a person when it should. The difference from a chatbot is what sits behind it: a general model with no access to your content will answer confidently and wrongly about your lead times and your terms, which is worse than no chat at all. A useful one is grounded in your documents, cites where an answer came from, refuses what it cannot support, and escalates anything worth money. Most of its value is earned outside office hours, on traffic you have already paid for.
Why we publish this
We do not publish client logos we cannot evidence, and a credentials deck proves nothing. What we can do is show our working — how we cost a build, what we monitor, where we would tell you not to spend.
Every failure mode described here is one we have had to fix, most of them on our own platform before any client saw them.
Where a number appears it is a formula you can put your own figures into, not an industry average from a slide.
A third of the automations we assess do not pay for themselves. Saying so publicly costs us projects and is the reason the rest are worth having.
Tell us the problem and we build a working demo on your own material within 48 hours. Free, yours to keep, and considerably more informative than any article.