Claude and GPT compete to design my Website
Want to back the stream? The «داعم» membership on YouTube is EGP 60 a month: badges, emoji, and I answer your questions first on stream. Become a supporter
The gist
The stream centers on preparing Burnish’s website as an early part of its go-to-market strategy: defining the brand essence and customer journey, communicating the product’s value within seconds, and testing landing-page messages and lead capture before the application is finished. Luqman practically compares Claude, GPT, and Codex for brainstorming, planning, and implementation, using visual references, screen recordings, and a context-aware judge; his preferences and impressions about model quality and efficiency are presented as personal observations, not general tests. The work includes reviewing the visual identity, palette, and temporary logo, balancing bold motion against clarity, and exploring Remotion and image-to-video workflows without documenting a completed final result. Audience discussions cover pricing, subscriptions, free offers, market selection, local models and tool security, and business ideas such as an AI trading bot, with emphasis on validating demand, margins, acquisition, and retention rather than merely building. A separate exploratory discussion considers Egypt’s El Dabaa nuclear power project under construction, its reported connection to the Russian company Rosatom, and a then-discussed target for the first of four plants or units to begin generating electricity in 2028; these details were mentioned in the conversation and not independently verified. Decisions about email gating versus deposits, some sales tools, the final design, and several technical experiments remain unresolved, while stream stutter and encoding lag were investigated without a confirmed fix.
78 moments from the stream
- 3:15:39
Diagnosing GPT Output and Comparing It with Claude
Luqman tried to diagnose a problem with GPT's output after a reference to verification through Chrome screenshots, but judged the visible result strange and unfinished despite the model reporting completion. In this particular run, he strongly preferred Claude's output and suggested continuing with it; the captions do not document a successful fix.
- 3:24:26
Adding a Dedicated Go-to-Market Workspace
A suggestion to add a marketing-focused area to a command center for organizing go-to-market strategy, acquisition pipelines, cohorts, interviews, and usage data according to the needs of each project.
- 3:26:19
Comparing GPT and Claude on Hero-Section Design
A practical comparison of two hero-section results produced from similar task handoffs. The speaker initially expected Claude to be far ahead, then revised that assessment after inspection: it was only somewhat better, with animation being the clearest visible difference rather than the expected dramatic leap.
- 3:28:15
Should Claude Plan for Codex, or Should Codex Do Both?
A qualified discussion of handing one model’s plan to an implementing model: the plan may work better when it states what to build rather than prescribing every implementation detail, while the speaker tentatively suggests that letting Codex reason and execute for itself could be preferable to splitting planning and execution across models.
- 3:36:20
Using a Reference Interface to Write Stronger Design Specs
After results described as bland and static, the speaker proposed studying a recording of Athena HQ’s interface and its related code, then reproducing the effects concept with a background suited to the target brand rather than copying all content literally. He later concluded that the main problem lay in the specifications and instructions more than in Codex’s implementation ability.
- 3:46:06
When Website Interactivity Becomes Too Much
Luqman reviewed highly interactive visual references and admired their quality, while arguing that game-like interaction can distract visitors and may be excessive depending on the product and its competitors. He noted that a simpler SaaS landing page can still be sufficient for its purpose.
- 3:50:20
Planning a Website Hero Video
He moved from an extreme interactive reference toward a more restrained idea: a looping hero video with an introductory or loading effect serving a specific purpose. He requested a complete prompt that could be passed to Codex instead of partial steps or deferred additions.
- 3:55:40
Creating Marketing Assets with Remotion in Codex
He described a workflow that starts with a new Codex session and precise instructions grounded in the project identity, logo, and description. He then presented Remotion as a tool that can be opened within Codex to edit assets and make marketing videos easier, while mentioning voice as an uncertain possibility.
- 3:57:27
Testing an Image-to-Animation Workflow
He planned to select a reference image, generate a landscape visual related to the project concept, and then animate it in Kling using start and end frames. He made a tentative comparison between Nano Banana and GPT, but the captions do not document a completed successful result.
- 4:08:23
Start With Bold Motion, Then Tone It Down
Luqman critiques an output that feels like a static image and asks for ambient motion, animation, and mouse interaction. He explains his design philosophy: first reach an intentionally aggressive version, then reduce motion or distraction if needed, because he believes AI recommendations tend to be conservative by default.
- 4:12:14
The Problem With Coding-Model Time Estimates
Luqman says an unclearly identified newer version of a coding model sometimes greatly overestimates task duration compared with an earlier version, based on his experience and impressions from people he asked. This is presented as a personal observation, not a systematic test.
- 4:13:57
Why AI Timeline Estimates Should Not Be Taken at Face Value
The speaker says the assistant’s duration estimates did not match the actual pace of implementation, so he removed the AI-generated timeline and reorganized the swimlane around domains.
- 4:15:29
Anchoring Website Design in Brand Essence
The speaker argues that website decisions should follow the brand essence, manifesto, and voice, since prescribing a visual element too early may conflict with the established brand direction.
- 4:26:18
Reviewing Decisions Across Implementation Phases
The passage outlines a workflow for reviewing decisions locked in an initial phase, checking whether they overlap with the next phase, and then discussing and finalizing the decision set before proceeding.
- 4:32:30
When Animation Obscures a Website’s Purpose
The speaker likes the site as visual inspiration but criticizes its heavy animation for making what it offers and why it exists unclear, highlighting the difference between technical showmanship and effective communication.
- 4:35:39
Balancing Animation and Clarity on Software Websites
The design is praised as visually impressive, while noting that extreme animation can confuse visitors and draw attention toward playing with the site instead of reaching the intended action or product. The expressed view is that software websites usually favor a less noisy approach.
- 4:38:56
A Clean Design and Hero Section
A brief positive reaction to a design direction that feels clean, with attention drawn to the page’s hero section.
- 4:42:14
Fitting Website Work into the Application-Building Workflow
He explains that the website is an important part of the workflow but not the core product, so he wants to complete it carefully and then return promptly to building the application and its remaining tasks. He also notes that web development is not his specialty and requires extra thought.
Got a question about something in this stream?
Ask in the group