Context Language Models
Agents organise their own working information to keep track of longer tasks.
0 → 1Earlier → later videosResearch ↗
YOUTUBE INTELLIGENCE · MONTHLY REPORT
What’s changing in AI?
New technical developments and changing discussion, drawn from our saved YouTube transcripts.
Report through31 Oct 2026Figures refreshed 1 Nov 2026
Videos in the archive8,332
Saved transcripts5,355
Channels represented920
Entire archive · Totals checked 1 Nov 2026 · Includes videos outside the comparison dates
Weekly refresh starts Monday at 8:00 am New Zealand time, using saved transcripts and the six latest completed calendar months.
Which topics are being discussed?
Matching videos · Share of analysed transcripts in brackets · Highest six-month total first
Click a count to see its videos. Hover over a topic, focus it with the keyboard, or tap it to see what it includes.
| Topic | Total3,091 analysed | Oct 2026968 analysed | Sep 2026585 analysed | Aug 2026378 analysed | Jul 2026359 analysed | Jun 2026397 analysed | May 2026404 analysed |
|---|---|---|---|---|---|---|---|
Agent teamsAgent teams Covers multi-agent systems, subagents, multiple agents, teams, swarms, parallel agents and orchestration: organising agent work. Subagents are included in this broad topic. The software factory wording alone does not qualify a video for this row. A video discussing both agent teams and software factories can appear in both rows. |
(13.6%) | (12.7%) | (10.8%) | (15.9%) | (13.1%) | (16.6%) | (14.9%) |
Coordinating agentsCoordinating agents Organising work between agents: multi-agent systems, teams, parallel agents and orchestration. Swarm wording alone belongs in Agent swarms. |
(7.1%) | (8.1%) | (5.5%) | (6.9%) | (4.7%) | (9.1%) | (7.7%) |
Reusable agent skillsReusable agent skills Reusable instructions and procedures packaged as agent skills, including SKILL.md files, custom skills, skill folders, packages, libraries, registries and standards. Also includes progressive disclosure: loading instructions when they are needed. |
(5.8%) | (4.5%) | (5.5%) | (6.9%) | (6.7%) | (5.3%) | (7.9%) |
Agent securityAgent security Prompt injection, agent security or identity, sandboxes and guardrails. |
(4.1%) | (5.2%) | (2.4%) | (3.4%) | (3.6%) | (5.8%) | (3.7%) |
Hosted and background agentsHosted and background agents Agents described as hosted, cloud-based, remote, asynchronous, persistent, always-on, background or long-running, including durable execution that preserves progress through interruptions. |
(3.8%) | (5.4%) | (4.6%) | (2.1%) | (2.2%) | (3.0%) | (3.0%) |
Computer useComputer use Explicit computer-use or computer-using wording. |
(2.6%) | (1.9%) | (5.8%) | (1.9%) | (3.1%) | (1.0%) | (1.5%) |
Human oversight and approvalsHuman oversight and approvals Human-in-the-loop work, human approval and approval gates. |
(2.4%) | (2.8%) | (1.9%) | (1.9%) | (2.2%) | (2.8%) | (2.7%) |
Context managementContext management Context engineering and management, Context Language Models, context compaction and context rot. |
(2.0%) | (1.7%) | (1.5%) | (1.1%) | (1.9%) | (2.5%) | (3.7%) |
Software factoriesSoftware factories Explicit discussion using the wording software factory or software factories. These describe workflows linking stages such as planning, coding, testing and review. This row is scored separately from agent teams; general discussion of multiple agents alone does not qualify. A video covering both topics is counted once in each row. |
(1.5%) | (2.5%) | (1.4%) | (2.1%) | (0.8%) | (0.5%) | (0.5%) |
Voice agentsVoice agents Voice agents, live voice and real-time voice. |
(1.2%) | (0.4%) | (3.1%) | (1.1%) | (1.9%) | (0.3%) | (0.5%) |
Knowledge graphsKnowledge graphs Explicit discussion of knowledge graphs as a way to organise information. |
(1.1%) | (0.2%) | (0.7%) | (0.5%) | (1.1%) | (3.0%) | (2.2%) |
Persistent memoryPersistent memory Persistent or long-term memory and memory banks. |
(1.0%) | (0.7%) | (0.7%) | (0.3%) | (1.1%) | (1.3%) | (2.7%) |
Self-improving agent systemsSelf-improving agent systems Discussion of self-improving or self-learning agents and recursive harness improvement, in connection with agent workflows, prompts, tools or software factories. This measures discussion of improvement methods; a matching video does not establish that a system improves reliably. |
(0.5%) | (0.6%) | (0.2%) | (0.8%) | (0.6%) | (0.8%) | (0.2%) |
Browser agents and automationBrowser agents and automation Browser agents and browser automation. |
(0.5%) | (0.4%) | (0.7%) | (1.1%) | (0.3%) | (0.3%) | 0 (0.0%) |
Agent swarmsAgent swarms Explicit discussion using the phrase agent swarm or agent swarms. |
(0.4%) | (0.6%) | (0.7%) | (0.5%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) |
Agent evaluationAgent evaluation Explicit agent-evaluation wording. This does not include every discussion of testing. |
(0.2%) | (0.1%) | 0 (0.0%) | 0 (0.0%) | (0.3%) | (0.5%) | (0.2%) |
Agent monitoringAgent monitoring Explicit agent-observability wording: observing agent behaviour. |
(0.2%) | 0 (0.0%) | 0 (0.0%) | (0.3%) | 0 (0.0%) | (1.0%) | 0 (0.0%) |
| Total topic matches | (47.9%) | (47.7%) | (45.3%) | (46.6%) | (43.7%) | (53.7%) | (51.5%) |
Topics overlap, so a video may appear in more than one row. Each video is counted once per matching topic. The bottom row totals topic matches, not distinct videos; its percentage may exceed 100%.
Each percentage uses that column’s analysed transcripts; the Total column uses all six months combined.
How these figures were calculated
The comparison uses publication dates in New Zealand time, the six latest completed calendar months and English transcripts containing at least 1,000 characters. Confirmed duplicate excerpts are excluded. Original topic groups require three qualifying phrase matches, or a matching title and one qualifying transcript phrase. Split rows require an explicit component phrase within a video that qualified for its original group, preserving proximity checks. A video can qualify for several components.
The vocabulary remains limited: agent evaluation requires explicit agent-evaluation wording, and agent monitoring requires agent-observability wording. These rows do not capture every discussion of testing or monitoring. Agent teams and software factories are separate topics; mentioning subagents alone does not qualify as coordination or swarms.
| Collection coverage | Oct 2026 | Sep 2026 | Aug 2026 | Jul 2026 | Jun 2026 | May 2026 |
|---|---|---|---|---|---|---|
| Recorded videos | 1,255 | 968 | 663 | 602 | 606 | 683 |
| Saved transcripts | 1,006 | 587 | 378 | 361 | 398 | 405 |
| Transcript availability | 80.2% | 60.6% | 57.0% | 60.0% | 65.7% | 59.3% |
| Transcripts analysed | 968 | 585 | 378 | 359 | 397 | 404 |
Shares describe discussion within this saved collection. The channel mix and transcript availability changed between periods; the figures do not measure wider industry adoption.
Technical developments to watch
Manually reviewed · Last reviewed 31 Oct 2026
Editorial comparison: 3 Aug 2026–13 Sep 2026 → 14 Sep 2026–25 Oct 2026. This section is updated after manual review.
Harness²
Agents test changes to their instructions, tools and methods, and retain useful lessons.
0 → 2Earlier → later videosResearch ↗
Knowledge Catalog
Agents identify gaps in shared business knowledge and propose corrections others can use.
0 → 1Earlier → later videosAnnouncement ↗
Counts refer to the editorial comparison dates shown in this section. Zero means no match in that sample.
ALSO THIS PERIOD
Agent Skills donated to the Agentic AI Foundation ↗ · Announced 22 Oct 2026. The donation is new; reusable skills were already established.
Watch the developments
Video publication dates · New Zealand time