Problems

Repositories are the evidence; the problems are the story. Crowded and rising is where the fight is — crowded and falling is a problem that was solved or abandoned.

Panel window January–July · refreshed monthly · last collected Aug 18, 2026

Problems below the significance floor — fewer than 40 people in February — are drawn hollow and sorted last. A move from eight people to nineteen is a handful of humans, not a trend.

All problems

Problem	Repos	February share	July share	Change
fleet-ops	78	3.6%	11.5%	3.22×
memory	105	6.7%	12.1%	1.80×
inference-cost	173	14.5%	22.4%	1.55×
token-optimization	35	4.0%	6.1%	1.51×
reliability	41	2.9%	4.4%	1.51×
agent-native-access	180	7.4%	10.9%	1.48×
evals	42	1.4%	2.0%	1.47×
observability	80	6.2%	7.8%	1.25×
supervision-burden	139	9.6%	11.5%	1.19×
data-analysis	48	2.2%	2.5%	1.14×
tool-use	179	10.8%	11.8%	1.10×
local-models	114	7.1%	7.3%	1.03×
config-transplant	81	7.6%	7.4%	0.97×
code-generation	212	29.2%	27.3%	0.93×
agent-safety	48	2.6%	2.4%	0.90×
context-engineering	141	9.7%	8.7%	0.89×
rag	92	5.1%	4.5%	0.88×
role-specialization	79	5.2%	4.6%	0.88×
agent-harness	425	60.2%	48.2%	0.80×
autonomous-research	46	2.0%	1.6%	0.79×
remote-control	52	4.3%	3.4%	0.79×
browser-control	39	1.3%	1.0%	0.77×
multi-runtime	181	34.1%	26.1%	0.76×
model-selection	1	1.2%	0.7%	0.56×
device-control	36	2.2%	1.1%	0.50×
discovery	23	2.2%	1.0%	0.47×
privacy	51	21.6%	6.7%	0.31×
code-reviewthin base	8	0.3%	1.0%	3.92×
tool-discoverythin base	24	0.6%	1.2%	2.12×
design-toolingthin base	54	0.9%	1.6%	1.77×
developer-experiencethin base	2	0.9%	1.2%	1.38×

To see the repositories behind any one of these, filter the Watchlist by problem.

These readings come from a rolling panel of 66,704 accounts, all polled across the full window; 20,330 recorded any activity inside it. How this is measured