Beamreach
First wave
Both people and companies are now developing custom software products from scratch. Products that integrate with their process, existing systems, etc. Scaling is going to be less important as those systems become essentially a ‘1 client saas’. Sometimes built in-house, sometimes by a vendor building just for you. Forward deployed engineers are exactly that.
In the first wave, this means purchases of application software will be close to 0. Not a lot of new licenses, and existing subscriptions will gradually be reduced in favour of in-house teams. But building gets cheap, owning doesn’t. Uptime, compliance, audits and liability cost the same. So we stop buying apps and keep buying infrastructure, data and accountability. Someone to blame is still worth paying for.
Some systems are easier to replicate, like a todo app for the team or internal messenger. Some are significantly stickier, especially anything related to security, payments, anything sensitive or critical. Industry shrinks rapidly before recalibration. Not a lot of jobs on the market. Startups are the only ones hiring at the moment. To be fair, rates and post-2021 overhiring are part of this too, not only AI.
Gradually, a lot of software professionals and newly graduated cs majors will become the ‘maintainers’. Lots of local systems, all need patching, testing, new features and edge cases handled. Some of the people currently working as managers are also going to turn into maintainers. Essentially, most of the employees will be contributing to developing local systems.
The winners of this wave will be (and frankly, already are) the fast adopters, the hosting and model providers, the select few ‘security oriented’ companies like cloudflare, palo alto, etc. Security is the obvious one. Thousands of bespoke systems, no shared patching. Nobody files a CVE for your internal app. And AI makes attacking bespoke targets cheap at scale.
One more thing grows quietly: the interfaces. When everyone’s internals are custom, the boundaries between companies become the shared layer. Protocols, data formats, standards. This is where companies first have to agree on what is true. Keep that in mind for the third wave.
Second wave
Second wave is maturity of those systems. Turns out with time they become harder to maintain, more day to day operations are required. Labor market turns favorable to employees again. New type of companies gain momentum: those who can produce highly flexible systems and those that build autonomous maintenance. Much of this grows together with the hosting/model providers.
There is a catch. If autonomous maintenance works, it eats the same jobs that made the market good again. My bet is complexity grows faster than autonomous maintenance matures. So there is a window. When it closes, the people in demand are the ones who can tell if the autonomous maintenance is right.
In this phase we sort of accept the elevated pace and adjust as professionals.
Third wave
Third wave is almost closing the circle. Systems become liquid around the core of truth. Code is a cache, regenerated when needed. What stays is the core: data, rules, invariants, tests.
The task of people is to “agree” on what is true.
This sounds far away, but I am already seeing it. AI-native teams describe their work like this:
“I go to a product meeting, we agree on a general direction, I receive an AI-generated jira ticket, my agent generates a PR, another agent reviews the PR, I glance at the results and merge.”
It’s somewhat anecdotal, but look at where the humans are. The meeting is the real work. That’s where people agree. Almost everything after it is generated, and the human just glances at the result.
And truth has layers:
- Facts. What happened. Ledgers, records, events. We agree through verification and audit.
- Rules. What should happen. Policies, contracts, regulation. We agree through negotiation and authority.
- Intent. What we want. Goals, priorities. We agree through leadership and, honestly, politics.
Those groups producing better truth get to survive. Better means more accurate, faster to update, and more accepted by the people who must act on it. Accurate but slow loses. Fast but contested loses too.
Why closing the circle? Before software, a business already was an agreement about truth. Double-entry bookkeeping, contracts, law. Software spent decades hard-coding those agreements. Now it goes back to interpreting them. Like a constitution: the core stays, the rulings keep changing.
New jobs: people who write specs, audit, arbitrate, know the domain, design tests. Verification becomes the main engineering skill.
Risks: whoever controls the core controls everything generated from it. Companies with incompatible truths can’t work together. And when disagreement is the bottleneck, office politics becomes engineering work.
Last note on timing. These waves are not taking turns. All of them are happening at the same time. AI-native teams are already touching the third wave. Some startups are in the second. Banks and hospitals haven’t started the first.
This is also how we think about it at Beamreach. Automate on what is certain, keep the rest liquid.
Related: Continuous Maintenance: the stage after CI/CD, the autonomous DevOps guide.