context window
architecture judged 75% likely a named technique or architecture
See how it has spread across media tiers
Mentions over time
Table view
| Week | Articles | Sources |
|---|---|---|
| 2023-W33 | 1 | 1 |
| 2023-W37 | 1 | 1 |
| 2023-W38 | 1 | 1 |
| 2024-W26 | 1 | 1 |
| 2024-W33 | 1 | 1 |
| 2025-W21 | 1 | 1 |
| 2025-W31 | 1 | 1 |
| 2025-W47 | 1 | 1 |
| 2025-W50 | 1 | 1 |
| 2025-W51 | 2 | 2 |
| 2025-W52 | 1 | 1 |
| 2026-W01 | 2 | 2 |
| 2026-W02 | 1 | 1 |
| 2026-W04 | 1 | 1 |
| 2026-W08 | 1 | 1 |
| 2026-W09 | 6 | 1 |
| 2026-W11 | 14 | 1 |
| 2026-W16 | 1 | 1 |
| 2026-W17 | 1 | 1 |
| 2026-W18 | 1 | 1 |
| 2026-W19 | 1 | 1 |
| 2026-W21 | 1 | 1 |
| 2026-W22 | 1 | 1 |
| 2026-W24 | 7 | 5 |
| 2026-W25 | 2 | 2 |
| 2026-W29 | 3 | 2 |
| 2026-W31 | 3 | 1 |
| 2026-W33 | 2 | 2 |
| 2026-W34 | 1 | 1 |
| 2026-W36 | 2 | 1 |
| 2026-W37 | 2 | 2 |
| 2026-W38 | 2 | 2 |
| 2026-W39 | 2 | 2 |
| 2026-W40 | 26 | 5 |
| 2026-W41 | 2 | 2 |
In their words
Multi-agent only becomes competitive when a single agent stops using its context window well, or when more compute is spent.
Kun Chen moves the threshold as the context window fills.
One context window holds the problem.
Even if you get through, you’re often left with a soup of nested <div> tags, telemetry scripts, and hidden navigation elements that eat up your LLM's context window and dilute the "signal" in your embeddings.
I would use a 128K total context window for common repair tasks.
Upstream API timeouts, rate-limit spikes, context window overflows, and malformed JSON outputs cause cascading UI crashes if unhandled.
ESAA-Conversational applies event sourcing to handoff across heterogeneous agents: a cold agent begins with handoff.md , state.md , decisions.md , tasks.json , and a selective context window projected deterministically from the event log.
I would also advise having a clear understanding of the physical limits of the model you are using: its context window, inference speed, and token consumption.