A Hertz executive asked to see the tablet view.
That’s the detail that stuck with me about Hertz’s 2019 suit against Accenture. The claim was that roughly $32 million had been spent, around $7 million of it on the discovery phase alone, for a device-agnostic responsive site with a reusable component library across several brands and countries. What was allegedly delivered was a North American site with security and performance problems and no reusable components. Five months past the deadline, with nothing finished, Accenture reportedly proposed a further $10 million to complete it.
And it was only discovered because somebody finally asked to look at it on a tablet.
I wrote about it at the time, and my conclusion probably wasn’t the easy popular one. Accenture deserved plenty of criticism. But Hertz had reportedly let go of its internal digital and development talent before the engagement began, effectively made its supplier the product owner, and then spent $32 million before anyone called a halt. My view was that client ownership is a, if not the, fundamental driver of a successful digital project, and that Hertz had put itself in an untenable position before anyone typed a line of code.
Before you file this under “enormous companies doing enormously stupid things”, though, it’s worth saying that the same failure at £150,000 with a five-person agency looks identical from the inside. It just doesn’t make the news.
Seven years on I keep thinking about the case, for a reason I didn’t anticipate.
It’s not a failure of procurement
Hertz failed because the company lacked internal capability. Management had removed the staff members who had the expertise to evaluate the work. No one could authoritatively state that the delivered product failed to meet the original requirements.
Once you’ve done that, the contract barely matters. You can have the tightest statement of work ever drafted (I’ve seen some very tight ones) and it won’t help you, because nobody on your side can tell whether it’s been met. Somebody eventually has to ask to see the tablet view, and by then you’ve spent the money.
These days we can realistically consider a team shipping code produced by agents, at volume, reviewed lightly because it mostly looks fine, in a codebase where the number of people who could confidently say why a given decision was made is falling.
That’s the same position. It just doesn’t require a supplier any more, unless you count an LLM as a supplier (perhaps we should).
What Hertz had that you don’t
Why am I writing this? Back then, Hertz could sue somebody.
There was a contract, a counterparty, a disclosure process, and a set of deliverables that either did or didn’t match what was ordered. Expensive, slow, and I imagine deeply unpleasant, but a mechanism.
Except look at what the mechanism produced.
Nothing. The case settled. It was voluntarily dismissed with prejudice and terminated in August 2021, terms undisclosed, and the only substantive ruling anyone got in two and a half years was a partial one throwing out a Florida consumer-protection claim. No finding on who owed what. No apportionment. One of the most visible public software delivery disputes of the time, $32 million on the table, and it ends with a confidential number and both parties agreeing never to mention it again.
That is the favorable result. In England, the outcome would be more limited. Your master services agreement likely limits the supplier’s liability to the fees you have already paid. This means the maximum recovery is a refund rather than full compensation for your losses. Legal costs are typically awarded to the winning party, so if you lose the case, you must pay for the other side’s lawyers. The process of disclosing evidence for a project that lasted several years takes a long time. This duration often exceeds the remaining time the companies I work with have before they run out of money.
The remedy was insufficient and was never used. This is significant because if an engineering organization loses the capacity to evaluate its own work, there is no remaining counterparty. There is no statement of work, no party to receive disclosures, and no information to disclose. This capability disappears gradually through staff departures and unreviewed pull requests. The loss is typically only identified when a failure occurs and no one can explain the original logic of the code.
Hertz bought a remedy it never got to use. You may not have one to buy.
I’m not arguing against using these tools. I use them daily and they’ve changed how I work. I’m arguing that the accountability question they raise is the Hertz question, and the Hertz question was never really about Accenture.
Two-way street, still
The other thing I said in 2019 was that a project of that size has to be a partnership with active day-to-day engagement, because that’s how misunderstandings get caught early rather than at the end.
I agree. This is why I am comfortable with agent-assisted development in teams that conduct proper reviews, but concerned when teams do not. The specific tool that wrote the code is not the primary issue. The important factor is whether a person is reading the output and has enough domain experience to identify errors.
Due diligence has caught up
There’s one genuinely new wrinkle, and it’s been showing up in my due diligence work.
When you review a codebase for an investor, you’re forming a view on whether the team can maintain and extend what they’ve built. That’s always involved reading code, but increasingly it involves asking about provenance. What proportion of this was generated? What review did it get? Can the team explain the parts that matter?
I don’t think there’s a settled answer yet on how to weigh that, and anyone claiming otherwise is guessing. But a codebase nobody on the team can explain is a liability whether a contractor wrote it, an agent wrote it, or a since-departed engineer wrote it in 2019. The mechanism differs but the risk doesn’t.
What I will say is that the question has got easier to ask and harder to dodge. Five years ago “who actually wrote this?” was close to an insult. It’s a normal part of the conversation now, and the teams who answer it well are noticeably calmer about it than the ones who’ve never thought about it. Which is itself a useful signal, and one that costs nothing to collect.
So, seven years on, would I put the blame anywhere different in the Hertz case?
No. And I’d add that the cheapest possible way to end up where Hertz ended up is to arrive there gradually, without ever signing anything.