Agentic Engineering
Running an org where the code has a coauthor that doesn't sleep.
Generation got cheap. That doesn't make your organisation faster on its own, it moves the constraint to review, judgment, and trust. This track is about running engineering where agents write a real share of the code, and building the org around what that actually requires.
What agentic actually means
"Agentic" doesn't mean a smarter autocomplete. It names a loop that plans, acts, and checks its own work without a person approving each step, and that loop is what changes both the value and the risk.
0-5 · 5-20 · 20-50 · 50-150 · 150+
Where agents earn their keep
The payoff is real in the parts of the job that are well-specified, bounded, and checkable. It's close to zero everywhere the actual work is judgment under ambiguity.
0-5 · 5-20 · 20-50 · 50-150 · 150+
Review capacity is the new bottleneck
Generation got cheap. If review, testing, and judgment don't scale with it, you've built a bigger queue, not a faster team.
0-5 · 5-20 · 20-50 · 50-150 · 150+
Autonomy levels and blast radius
The permission an agent has, not the model behind it, is the real governance question. Decide autonomy per task and its reversibility, not once for the whole tool.
0-5 · 5-20 · 20-50 · 50-150 · 150+
Evals, not vibes
"It feels faster" is not a measurement. The organisations getting real value from this are the ones that measured something specific before and after.
0-5 · 5-20 · 20-50 · 50-150 · 150+
What junior engineers should be learning now
The apprenticeship that built engineering judgment ran on writing a lot of mediocre code and getting it corrected. That loop is gone, and it didn't need rebuilding so much as replacing: juniors who review more than they write, and ship sooner, grow faster than the old one ever managed. The failure isn't the agent writing the code. It's the junior not reading it.
0-5 · 5-20 · 20-50 · 50-150 · 150+
Build vs buy for agent infrastructure
Buy the model, buy the harness. Build the context and the guardrails. Most of what actually differentiates your setup, and most of the risk, lives in those last two.
0-5 · 5-20 · 20-50 · 50-150 · 150+
The spec is what you review now
For well-specified work, the spec is what gets approved, not the diff behind it. Get the spec wrong and you've automated the mistake instead of catching it.
0-5 · 5-20 · 20-50 · 50-150 · 150+
Context is maintenance, not a build
Agent context files don't rot because nobody cares. They rot because updating them was never anyone's job. The team that changes the behaviour maintains the file, and one named leader is accountable for whether it still tells the truth.
0-5 · 5-20 · 20-50 · 50-150 · 150+