Why ChatGPT and Claude keep disappointing you

The AI Corner, by Ruben Dominguez

---

1. Key Themes

The Real Bottleneck Is Configuration, Not the Model

Most users blame poor outputs on the AI itself, but the article argues the gap is entirely in setup. Persistent context features — Projects, Memory, Custom Instructions — go undiscovered by the majority of users, meaning they restart from zero every session.

"Every session starts fresh. The tool carries zero memory of who you are, what you are building, or how you like to communicate. It defaults to something generic, and you spend time reshaping the output before it is usable."

The Power User Gap Is a Feature Awareness Problem, Not a Skill Problem

The delta between average and advanced AI users is not intelligence or prompting sophistication — it is knowledge of which features exist and how to turn them on.

"The gap between a basic user and a power user is not intelligence or technical skill. It is knowing which features exist and how to activate them."

Free vs. Paid AI Tiers Represent a Massive Capability Cliff

The article frames the free-to-paid upgrade not as incremental but as a categorical difference in what the tool can accomplish, citing benchmark data to substantiate the claim.

"Over 900 million weekly ChatGPT users, and the vast majority rely on the free version, which lacks advanced reasoning and planning. The thinking model available through subscription is tied for first in performance benchmarks. The free version ranks 25th."

AI Tools Are Evolving Into Persistent, Context-Aware Colleagues

The article frames the end-state of proper setup not as a smarter chatbot, but as something qualitatively different — an always-on collaborator with accumulated institutional knowledge of the user.

"Together, these features turn a chatbot into something that feels like a colleague who has been working with you for months, one who already knows your priorities, your voice, and your current projects before you type a single word."

---

2. Contrarian Perspectives

Better Prompts Are the Wrong Lever — System Setup Is What Actually Moves the Needle

The consensus among casual AI users is that writing better prompts unlocks better results. The article directly challenges this, arguing that prompt optimization is largely wasted effort without foundational configuration in place first.

"Dramatically better results from both tools are available without changing a single prompt. You just have to set them up properly first." "The setup is the variable that changes everything, not the prompts."

Most Users Are Operating at ~20% of Claude's Actual Capability

The common assumption is that opening Claude in a browser tab and typing a query gets you "Claude." The article argues this delivers only a fraction of the tool's true output capacity — the rest is locked behind the desktop app and its Cowork feature.

"Browser-tab Claude delivers roughly 20% of what the tool can do. Cowork on the desktop app is where the other 80% lives."

---

3. Companies Identified

---

4. People Identified

---

5. Operating Insights

Build a One-Time System, Then Collect Compounding Returns Daily

The article's core operating thesis is that AI setup is a one-time investment with indefinite daily payoff. Operators and entrepreneurs should treat AI configuration the same way they treat SOPs — build it once, benefit continuously.

"The people getting genuinely different results built a system around the tool once and have been collecting the returns every day since."

Use Two Core Files in Claude to Replace Dozens of Repeated Prompts

Rather than re-contextualizing Claude in every session, the article prescribes two persistent reference files — about-me.md (personal/professional context) and anti-ai-style.md (voice and tone rules) — that auto-load and effectively eliminate repetitive prompt overhead.

"The two core files — about-me.md and anti-ai-style.md — that replace 50 prompts."

Use Scheduled Tasks to Extend AI Productivity Beyond Active Work Hours

Both tools support workflows that execute asynchronously, enabling operators to queue work from their phones and have outputs ready without manual supervision.

"Scheduled tasks: recurring workflows that execute while you are offline." / "Dispatch: how to queue tasks from your phone while your computer does the work."

---

6. Overlooked Insights

Connectors Enable Live Integration Across the Entire Modern Work Stack

The article briefly lists Connectors — live access to Slack, Google Drive, Notion, Gmail, and 50+ other tools — as a feature most users have never activated. For operators already running on these platforms, this represents an immediately available integration layer that most are leaving dormant, with no third-party automation tool required.

"Connectors: live access to Slack, Drive, Notion, Gmail, and 50+ other tools."

Custom GPTs Are a Team-Level Force Multiplier, Not Just a Personal Tool

The article mentions the ability to build Custom GPTs encoded with a specific voice and ruleset that are shareable with a team — a lightly noted point with significant implications for standardizing AI output quality across an organization.

"Custom GPT builder guide: your voice, your rules, shareable with your team."