Before AI, not much.
Today, a good engineer with AI is starting to look a lot like a cyborg engineer: same human judgment, experience, and accountability, but augmented with tools that can write code, analyze systems, search documentation, review configurations, and test ideas at a speed that wasn't possible before.
AI gives an engineer more power.
So naturally, some companies look at that and think:
"Great. Maybe we can get rid of the engineer."
And replace them with what, exactly? An AI agent? A collection of agents? Eventually a robot?
I don't think that ends well.
The engineer understands why the system exists, what can break, what the business actually needs, and — critically — when the AI is confidently producing nonsense. That last one is not a hypothetical. It's the daily reality of working with large language models. Without a human in the loop who can recognize a plausible-sounding hallucination, the system will eventually ship one into production.
AI doesn't eliminate engineering responsibility. It makes a capable engineer dramatically more productive.
Concretely, that means:
- The engineer still owns the design.
- The engineer still owns the blast radius.
- The engineer still owns the call when the model is wrong.
What changes is the unit economics of attention. Boilerplate, lookup, and first-draft work — the kind of thing that used to eat half a sprint — can now be offloaded. The human time that remains gets spent on the parts only a human can do: framing the problem, weighing trade-offs, and judging what "correct" actually means for this codebase, this team, this customer.
There is another part of this conversation that doesn't get enough attention: cost.
Right now, AI can feel incredibly cheap. Subscriptions. Subsidized plans. Bundled tokens. Aggressive competition between providers. The sticker price of "AI for everyone" looks like a rounding error next to a senior engineer's salary.
But look at the underlying usage. If you run serious agentic workflows and trace the equivalent API consumption through observability tools such as Langfuse, seeing hundreds of dollars per day in model usage is not difficult. I've personally watched workloads run in the $300–$400/day range when an agent is doing real work — reading repos, calling tools, iterating on failures, re-reading context.
That is not a typo. That is one engineer with one AI producing four figures a month in token spend.
At some point, somebody pays that bill. And if the answer to "who pays?" is "the model provider's investor", then the math that justifies replacing senior engineers is built on a price that won't last.
The winning model, in my view, is simpler than the discourse suggests:
- Good engineer + AI > good engineer alone.
- But: AI ≠ engineer.
Use AI to build cyborg engineers. Engineers whose instincts, experience, and accountability are augmented with tools that move faster than any one human could. Engineers who can ship more, see more, and learn more per hour than they could last year.
Don't use AI as an excuse to remove the humans who understand what they're building. The short-term savings will look real. The long-term bill — measured in outages, in security incidents, in software nobody trusts — will be paid by someone. It usually isn't the executive who approved the swap.
Cyborgs and engineers have more in common than the org chart suggests. They are both systems where the irreplaceable part is not the metal or the model — it's the judgment of the operator. Build for that. Hire for that. Augment that.
The future of engineering isn't fewer engineers with more agents. It's better engineers with better tools.
#AI #Engineering #SoftwareEngineering #DevOps #SRE #ArtificialIntelligence #AIAgents #FutureOfWork #Technology #SoftwareDevelopment