Sideways is the New Up
Read the Blueprint →Ronnie Phala · June 2026
The first great mistake of the AI era may be that companies are letting go of the very people they need most.
Across boardrooms, executives are asking a simple question: if AI can do 70% of a person's job, what happens to the person?
The answer so far has often been a redundancy notice. But the evidence is beginning to suggest this may become one of the most expensive mistakes of the transition.
The early lesson came from Klarna. It celebrated an AI assistant performing the work of 700 customer service agents. Then it changed course. Its CEO later admitted that the pursuit of lower costs had gone too far and quality had suffered.
The pattern became common enough to earn a name: the AI boomerang.
Companies cut people after automating work, only to discover they had also removed something the software could not replace: context, judgment, relationships, and the practical knowledge of how work actually gets done.
Many are now hiring back, often at a premium. Research from Gartner, Forrester and Robert Half has tracked the same pattern, with a large share of firms rehiring for roles they had only just cut.
This is not a story about AI failing.
The technology did automate the tasks.
It is a story about companies discovering, the expensive way, that tasks and work are not the same thing.
The Talent Contradiction
At exactly the same moment companies were letting these people go, the market developed an enormous appetite for a newly scarce profile: the Forward Deployed Engineer (FDE). The FDE is someone who can enter a messy organisation, understand its workflows, and turn AI capability into production systems.
AI labs and investors have now institutionalised this model. OpenAI's roughly $4 billion Deployment Company and Anthropic's $1.5 billion deployment venture are both built around scaling forward deployed engineers into enterprises.
The smartest money in AI has identified the implementation bottleneck. What remains unclear is whether companies need to buy all of that capability from the market, or whether part of it is already emerging inside their own workforce.
The contradiction is stark.
Companies are paying a fortune to import outsiders who have AI implementation skill but no institutional knowledge, while laying off insiders who have acquired both.
In the process, they may be discarding a capability that the market increasingly treats as scarce and valuable.
The Invisible Third Category of Talent
Every automated workflow frees a person. The real managerial question is what happens next?
Fire them and you risk the boomerang. Keep them exactly where they are and you have unused capacity.
The deeper problem is that our model of talent only has room for two types of people.
The internal operator knows the systems, the history, the politics, and the hidden complexity of the business, but rarely has the time, training, or mandate to redesign it.
The external consultant or engineer arrives with implementation skill and authority, but is expensive, transient, and still learning how the organisation actually works.
Neither represents the bridge this moment requires.
A third category is emerging organically from the AI transition, and the organisations that recognise it, define it, and build around it will be the ones that capture its value.
The irony is that organisations already on the AI transformation journey are producing these people themselves. What is missing is the structure and mandate that allow them to be discovered, formally categorised, and deployed.
We call this emerging category Sideways Deployed Talent.
The full architecture, governance model, economics, and operating design are outlined in the SDS Blueprint →
The Sideways Deployed Specialist
The Sideways Deployed Specialist (SDS) is an insider redeployed sideways across an organisation or portfolio as a peer-level AI implementer.
It is what you'd get if a Forward Deployed Engineer and a seasoned super user had a child.
From the engineer comes AI fluency, implementation discipline, and the ability to ship solutions into real-world complexity.
From the super user comes institutional memory, domain expertise, and peer-level trust.
The SDS borrows from both while avoiding the limitations of each.
These are practitioners who have already automated large parts of their own work using AI agents. In doing so, they have acquired something rare: firsthand knowledge of what it actually takes to make AI work inside a real organisation.
They are then trained in implementation, advisory, and change capability, and deployed sideways into other teams with support from AI tools and a small central capability that captures and spreads what they learn.
The breakthrough is not simply productivity.
It is organisational legibility.
Every time an SDS redesigns a workflow, hidden knowledge becomes visible because you cannot automate a process you cannot explain.
Deploy enough SDSs across an organisation and it begins to understand itself.
People Must Be Safe to Be Found
There is one problem that every AI transition strategy must solve.
A rational employee who automates 70% of their job does not announce it. Why reveal the very evidence that could be used to make you redundant?
Any serious AI transition model must reverse that incentive.
Finding and revealing opportunities for automation must become the gateway to a new career path and greater value, not the beginning of an exit process.
Why Champions Are Not Enough
By now, a reasonable question emerges: isn't this just a change champion with a better title?
Not quite.
Change champions proved the principle but not the model.
We ask champions to lead a transformation with a volunteer's toolkit and barely any time to use it. So the day job wins, and the transformation stalls. And nothing holds them once they master the one skill the market is bidding up.
The SDS automated their own job, and earned three things at once: spare hours, rare AI skill, and a misplaced target on their back.
The SDS model takes the same peer-to-peer mechanism and gives it the training, mandate, incentives, and infrastructure needed to operate at scale.
The champion was the prototype.
The SDS is the scaled production system for the AI era.
Policy Buys Time. Companies Must Decide What to Do With It.
Governments can soften the impact of AI displacement. They can create incentives, insurance mechanisms, and transition frameworks.
But layoffs happen one organisation at a time, in management meetings, not legislative chambers.
The SDS model offers a practical answer.
It turns people displaced by AI into the people who spread AI throughout the ecosystem.
It does so using economics that executives can understand: redeploying salary that is already being paid instead of constantly buying external consulting capacity.
The model also points to a gap in the market.
The AI labs have built tools for marketing, law, software engineering, finance, and research.
They have not yet built tools for the transition itself.
The next great opportunity may lie in building vertical AI tools that help organisations bridge the transition and deploy this emerging talent category at scale.
The scarcest resource of the AI era is not capital.
It is not compute.
It is people who understand how work actually gets done and have demonstrated they can redesign it.
This new talent category is already forming in plain sight. The winners of the AI transition will be the ones who recognise it and harness its value before everyone else does.
The SDS Blueprint — architecture, governance, economics, and tools — is open and free to run →
Read it. Test it. Improve it. Report back, including the successes and challenges.
Sideways is the new up.