Travel brokers took 10 years to break down. Developers are 3 years in.

Travel brokers are the go-to instance of an trade killed by the web. And the numbers are brutal: US brokers peaked at round 124,000 in 2000 and collapsed to underneath 40,000 by 2020 – a 70% drop. Retail places fell from 34,000 to 13,000. But that collapse took a decade. The ones who survived did it by going upmarket. I hold enthusiastic about this once I take a look at what’s coming for software program engineering – besides this time, I do not assume we get ten years.

The historical past

Interestingly, whereas researching this text there are some important different components at play. US airways dramatically reduce commissions in 1995 – which accounted for 60% of the common US journey agent’s income previous to this.

While it is onerous to say how a lot of this fee reduce was as a result of growing digitisation of the reserving companies vs different components, I really feel this actually does have parallels to the software program engineering market, which had an enormous boon with covid-era ZIRP inflicting arguably far an excessive amount of capital to be allotted to software program engineering after which a gradual however seemingly relentless pullback in job positions submit that.

In some ways the fee reduce led the journey agent trade to be within the worst potential place for the arrival of the web with critical virtually in a single day cashflow worries. I’ve little doubt this led the trade to be poorly ready for the arguably a lot bigger risk of OTAs – margins actually began eroding however general journey volumes continued to extend, masking the structural shift occurring.

I do really feel that is occurring with software program engineering positions and contracts proper now. Many chats I’ve with folks appear in charge the economic system or different exterior components for the massive slowdown in openings. While there’s some reality on this – there was a ~$150bn decline in US VC funding – I additionally hear lots of managers and CTOs saying they aren’t needing to rent for extra software program engineering positions and infrequently aren’t rehiring when staff depart.

Things received a bit higher, then received lots worse

Interestingly, employment within the US journey agent sector began to enhance within the late 90s – as a result of report journey volumes. It was a traditional ‘make it up in quantity’ play, the place margins began eroding due to the fee reduce. Anecdotally I’ve heard of lots of this occurring within the customized software program engineering market – with important discounting occurring to try to keep/enhance headline income numbers albeit at a (a lot) lowered margin.

It’s essential to bear in mind in 1999 lower than 5% of journey was booked on-line – which appears extremely alien to us now.

LLMs have gotten far extra market share in far much less time. This is the important thing cause I feel the modifications are going to be way more fast. We’re at ~2.5 years for the reason that launch of GPT-4 (the primary mannequin that would actually try and code on any critical degree) and LLM utilization is >40% of the whole US inhabitants.

Even extra astoundingly, in keeping with the Stack Overflow developer survey LLM adoption in software program engineering went from 0% in 2022 to 84% (!) in 2025.

Who survived

Interestingly, whereas the market contracted quickly with OTAs seeing very fast development over the early 2000s, there have been some markets that noticed main development.

Corporate “TMCs” (journey administration firms) noticed big development – the businesses in control of mass-booking worker journey on behalf of firms.

So did sure area of interest elements of the market – cruises particularly (nonetheless 75% offline). Luxury journey exploded – Virtuoso up 211% – arguably as a result of they’re accessing stock that is not out there to anybody.

So it wasn’t all unhealthy information. There was definitely some resilience within the journey trade the place there was extra complexity, sometimes requiring a number of merchandise packaged along with greater commissions on some merchandise outweighing the wafer skinny (or non-existent) commissions on airfare.

Who did not

Generalist journey brokers received fully worn out. Retail journey company institutions fell 59% between 1997 and 2013; from almost 23,000 to underneath 10,000. If your job was to sort buyer necessities into Sabre, inside just a few years you have been competing immediately with an internet site that would do it quicker and cheaper. The most commoditised work went first: easy point-to-point flights moved on-line virtually instantly, and by 2002 brokers who trusted airline ticketing had zero fee and no differentiation.

Between 2000 and 2020, someplace between 58,000 and 64,000 brokers exited the career fully. There was no retraining program. Most did not “move upmarket”.

I feel it is a very telling story for software program engineering. If your job is to translate necessities into code manually – and that is it – you are the generalist journey agent.

I’m nonetheless talking to far too many software program engineers who’re dismissive of agentic tooling, or who deal with it as a novelty slightly than the factor that is coming for his or her job. If you are preventing it slightly than leaning in, until you are fortunate sufficient to be in a particular area of interest, I think the market goes to look extraordinarily ropey over the following 5 years.

That’s to not say that software program engineering is “done” – removed from it. Some of the very best engineers I do know have leveraged it to enhance high quality whereas growing productiveness. For instance – constructing higher check suites, higher observability and likewise prototype a number of instructions to see what works finest. Or they’ve massively improved the standard of the UI/UX for MVPs in the event that they lean backend.

Software engineering does not have 10 years

As the stats above present, adoption is extraordinarily fast. The different curve that’s occurring from my final submit which blew me away is the advance in agent success charges from METR.

Opus 4.5 has actually startled me – it genuinely can do complicated software program engineering duties which I’d count on a proficient developer to take hours in minutes with only a few defects.

The actual query is what occurs over 2026. Are we going to see ‘superhuman’ brokers that far surpass human talents (both in pace, high quality or another dimension we have not even considered)? I do not know. But I’m not ready round to search out out.

What does ‘upmarket’ seem like?

The actual worth now lies in area data: understanding how techniques join, realizing which information exists the place, and greedy what the enterprise really wants. I’ve had outrageously good outcomes taking my data of inside and exterior information sources and getting LLM brokers to synthesise all of it collectively. That sort of work is not going away: if something, enhancements in agentic coding imply you are able to do what would have required a crew of 10 in what looks like just a few afternoons.

The different transfer is to broaden. If you are a backend engineer who’s at all times prevented frontend, now’s the time – brokers can bridge the hole when you study. If you are frontend-only, lean into backend, devops, infrastructure. The engineers I see thriving are those who can personal a whole downside end-to-end, not simply their slice of it. The generalist journey brokers received worn out, however the generalist engineers – those who can transfer throughout the stack – are extra priceless than ever.

Travel brokers had ten years to determine this out. Most did not. Developers are three years in, and the curve is steeper.