Are AI agents changing software development?
An editorial case on how agentic coding tools are reshaping development workflows — what is observable, what is hype, and which questions remain open.
01 · Observe
WHAT DO WE SEE?
- Agentic coding tools that plan, edit, test, and iterate are now widely available to working developers.
- Teams report meaningful speedups on boilerplate, scaffolding, and well-specified tasks.
- Complex, ambiguous, and architecture-level work still visibly requires experienced human judgment.
02 · Evidence
WHAT SUPPORTS IT?
Tool adoption reports
Industry surveys and vendor reports describe rapid uptake of AI coding assistants. Figures vary by source and methodology — no single number is asserted here.
Source: Industry surveys and vendor reports (editorial reference)
Practitioner accounts
Working developers widely describe productivity gains on routine tasks alongside new failure modes (plausible-but-wrong code, review burden).
Source: Practitioner writing and conference talks (editorial reference)
03 · Context
WHAT SURROUNDS IT?
- Every automation wave in software (compilers, frameworks, cloud) changed workflows without eliminating developers.
- Vendor incentives favour strong capability claims; independent measurement lags adoption.
04 · Perspectives
HOW ELSE COULD IT BE READ?
The Practitioner
Faster on the routine, same on the hard
Day to day, agents compress the boring parts. The bottleneck moved: specifying well and reviewing carefully matter more than typing speed.
The Sceptic
Measure, don't marvel
Most evidence is anecdotal or vendor-supplied. Until independent studies track defect rates and long-term maintainability, treat productivity claims as hypotheses.
05 · Hypotheses
WHAT COULD IT MEAN?
Agents raise developer throughput on well-specified work
PARTIALLY SUPPORTEDFor tasks with clear specs and fast verification, agentic tools produce net speedups.
Supporting
- + Broad practitioner agreement on routine-task acceleration.
Contradicting
- − Independent controlled measurements remain scarce.
Agents reduce the need for junior developers
OPENEntry-level implementation work is increasingly automated away.
Supporting
- + Some hiring commentary points in this direction.
Contradicting
- − Review, specification, and domain judgment still require humans learning the craft somewhere.
- − Historical automation waves shifted junior roles rather than removing them.
Open hypotheses
Competing hypotheses stay open and converge only where evidence allows. Unresolved tension is a result, not a failure. Node positions carry no quantitative meaning — only the status labels matter.
06 · Challenges
WHAT COULD BE WRONG?
Challenges: Agents raise developer throughput on well-specified work
Throughput of what — lines shipped, or working systems maintained over time? What happens to defect rates?
07 · Open questions
WHAT REMAINS UNRESOLVED?
- What do defect rates look like in agent-assisted codebases after twelve months?
- Which verification practices actually catch agent-introduced errors?
08 · Reasoning
HOW DID WE GET HERE?
DAMNO AI — Reasoning
WHAT WE KNOW
- Agentic coding tools are widely available and adopted.
- Practitioners report routine-task speedups.
- Independent controlled evidence remains limited.
WHAT WE INFER
- The bottleneck is shifting from writing code to specifying and verifying it.
- Vendor-supplied evidence should be discounted relative to independent measurement.
WHAT WE ARE TESTING
- Throughput gains on well-specified work (partially supported).
- Junior-role displacement (open).
WHAT REMAINS UNCERTAIN
- Long-term effects on code quality and maintainability.
- How the junior-to-senior pipeline adapts.
No approved contributions yet. Share a perspective, question, hypothesis, evidence or challenge below to enrich this reasoning.
09 · Collective
HOW UNDERSTANDING IS FORMED.
COLLECTIVE REASONING
The current understanding is based primarily on curated research.
As contributions are added, new perspectives, challenges and evidence will become traceable here.
09 · Collective
HOW UNDERSTANDING IS FORMED.
COLLECTIVE REASONING
The current understanding is based primarily on curated research.
As contributions are added, new perspectives, challenges and evidence will become traceable here.
09 · Collective
HOW UNDERSTANDING IS FORMED.
COLLECTIVE REASONING
The current understanding is based primarily on curated research.
As contributions are added, new perspectives, challenges and evidence will become traceable here.
09 · Learning
WHAT DO WE TAKE FORWARD?
- Separate adoption (observable) from impact (needs measurement).
- Shift the question from 'will developers disappear' to 'where does the bottleneck move'.
10 · Contribute
ADD A PERSPECTIVE.
The conversation is open.
No approved contributions yet. Share the first perspective, question, hypothesis, evidence or challenge below.