Summary of "I Think They Are Lying To You"

Overview

The subtitles present an argument that challenges claims that AI coding assistants have “solved” coding. Instead, it argues that product limitations and reliability/UX issues remain significant and are being oversold.

Core technological/product claims discussed

Shift from “prompting” to “looping” with Claude Code

Boris (Anthropic) is quoted as saying he uninstalled his IDE and no longer “prompts Claude” manually.
Rather than direct interaction, he describes using automation via loops:
- run iterative processes until a win condition is met
- “burn” tokens as needed during the loop
The framing is that this represents an evolution from interactive prompting toward agentic workflow loops.

Company/market framing: “coding is the easy part” / “coding is solved”

The speaker argues that this kind of messaging implies humans don’t need to write code directly anymore.
Instead, humans supposedly just configure loops and rely on automation.
A cited Anthropic acceleration claim is 8× more code per employee per quarter (relative to pre-2025 averages), used to support the “coding is easy/solved” interpretation.

Main critique: reliability and UX issues contradict “coding is solved”

Claude Code “terminal flickering” bug

Claude Code (terminal app) released for researchers in February 2025.
The speaker claims a terminal rendering/text flickering bug was reported within two weeks and persisted for a long time.
Anthropic response (as cited) claims they rewrote the terminal rendering system to reduce flickering by ~85% (reported as of December 17, 2025).
The speaker emphasizes the mismatch between:
- “coding is solved” marketing/messaging, and
- the prolonged existence of a software-rendering issue (argued to be software-related, not hardware/capacity).
Further claims by the speaker:
- Even after a long duration, flickering allegedly wasn’t fully resolved.
- A “no flicker mode” was released by Boris/Claude Code on April 1 (speaker claims it uses alternate screen, similar to Vim, helping avoid direct printing that’s harder to render reliably).

Ongoing error messaging / unclear failures

The speaker points to Claude Dev’s official updates (ending with May 27) and criticizes error handling:
- “mysterious” errors rather than clear explanations
- examples implied include:
  - connection terminated
  - out of credits
  - inability to continue
The critique is that, in these cases, the system should loop or handle issues more transparently.

Claims of persistent model/service issues

The speaker alleges:

Elevated “errors” visible on Claude status, while still citing roughly 98% uptime as errors reportedly continue.
Incidents where users claim they received other people’s feedback/prompt results, leading to confusion about unexpected outputs (including legal documents).
The speaker notes that this is presented as possible session leakage, but also states they do not definitively claim leakage is proven.
Their argument is that Anthropic “can’t really solve” or clearly explain the issue, and that it has persisted.

Claimed human impact

The speaker argues the messaging causes harm:

Burnout and fear among developers who believe they must “yeet code into production” with insufficient controls.
Concern that people may be encouraged to rely on token-heavy autonomous loops without adequate safety/verification.

Overall conclusion of the speaker

The speaker’s thesis is that the “coding is solved” narrative is misleading:

either intentional (“lying”), or
self-deception despite evidence to the contrary.

They acknowledge AI can accelerate code production and automation, but argue the gap between public claims and real-world reliability/UX problems is too large.

Main speakers/sources mentioned

Boris — identified as Anthropic (creator of “Claude Code,” presented as the main quoted voice).
Anthropic / Claude Dev official account — cited for product communications (terminal/flicker fixes, rendering updates, and error/status-related messaging).
GitHub issues / Reddit / Hacker News posts / Claude status page — referenced as sources of reported problems and user allegations.