Show the evidence.
Link to studies, protocols, code or first-party documentation; expose sampling limits and conflicting results.
INDEPENDENT / EVIDENCE-LED / OPEN TO REVISIONEDITION 01 — OCT 2026
Tomorrow isn't
a conclusion.
AI-assisted programming is changing the way software gets made. We track the evidence, test the claims and keep the forecasts accountable.
THE OBSERVATORY'S POSITIONCuriosity without hype. Skepticism without dismissal.
01OBSERVATIONWhat a source actually reports
02INTERPRETATIONWhat the evidence may imply
03FORECASTWhat has yet to be demonstrated
WHY AN OBSERVATORY?
More signal.
Less certainty theater.
Emerging software tools are presented through benchmarks, demos, survey responses and confident predictions. None of these sources answers every question. Our job is to identify what each can—and cannot—support.
Read the evidence protocol ↗46%of responding developers in Stack Overflow's 2025 survey distrusted AI-tool accuracy.
Survey evidence is a snapshot of respondents, not a census of all developers. Examine the primary results ↗
FIELD INDEX / 10 DOSSIERS
Where the next era
is being examined.
Ten perspectives. Separate claims. A common standard of traceable evidence.
⌕
01 / DOSSIER
Emerging Paradigms
New modes of authoring, assembling and maintaining software.
Explore research ↗02 / DOSSIERVibe Coding
Programming through prompts, generated code and iterative feedback.
Explore research ↗03 / DOSSIERHuman–AI Collaboration
New divisions of labor between developers, reviewers and models.
Explore research ↗04 / DOSSIERAI-Native Development Tools
Editors, command-line agents and integrated coding workspaces.
Explore research ↗05 / DOSSIERBeyond Text-Based Coding
Visual, conversational, spatial and multimodal development interfaces.
Explore research ↗06 / DOSSIERTesting & Verification
Benchmarks, correctness, security and the limits of passing tests.
Explore research ↗07 / DOSSIERProgramming Education
What should developers learn when code can be generated?.
Explore research ↗08 / DOSSIERSoftware Economics
Productivity, costs, maintenance and organizational change.
Explore research ↗09 / DOSSIERFutures Observatory
Scenarios, indicators and claims that can be tested over time.
Explore research ↗10 / DOSSIERExperiments & Essays
Replicable trials, methodological notes and critical perspectives.
Explore research ↗No matching research sections. Try a broader term.
FEATURED RESEARCH / EVIDENCE REVIEW 001Does AI coding
Does AI coding
actually make us faster?
One controlled trial measured slower completion. Newer investigations raise different questions. We examine why task selection, self-reports, and benchmark validity complicate the headline.
Read the evidence review ↗EARLY-2025 METR TRIAL+19%time taken with AI in the studied setting
LATER REASSESSMENTUNRESOLVED2026 follow-up flagged selection effects
Numbers are from particular studies, not universal estimates. Read METR's update ↗
PUBLIC RESEARCH CONTRACT
A prediction can be wrong.
Its record shouldn't disappear.
State the uncertainty.
Separate observations, hypotheses and measurable forecasts. Record assumptions before outcomes are known.
Make updates visible.
Invite counterevidence, acknowledge corrections, and maintain a consequential public revision history.
What will programming
become?
Evidence first. Futures second.
ENTER THE OBSERVATORY ↗