Can we replace a $150k Looker with a bucket of HTML files?
david.lequin.dev ·
A three-engineer data platform team at a roughly 500-person company built an internal AI-generated dashboard service that hosts HTML files, runs live Snowflake queries, and has reached at least as many weekly users as the company’s Looker deployment. The company pays about $150,000 annually for roughly 300 Looker seats, while the internal service grew from 23 weekly viewers in March to 136 in June. Users generate free-form dashboard pages with Claude, publish them through an internal connector, and share them behind single sign-on.
The service initially exposed serious security risks because unreviewed JavaScript ran on the same origin as user sessions and every employee could access every dashboard. The team later added a remote MCP connector, draft-and-publish workflow, sandboxed iframes on separate origins, strict content-security policies, one-hour viewer-specific content tokens, and metadata scanning that can flag or block risky scripts and WebRTC connections. Snowflake OAuth supplies each viewer’s identity and data permissions, so sharing a dashboard does not grant access to underlying data.
The team reports 325 commits, about 74,000 lines of code including 20,000 test lines, and roughly $6,400 in Claude token costs over eight months. It estimates annual operating costs at about $20,000, including AWS and ongoing Claude use. Its analyst agent uses about 1 million tokens per dashboard, with reported costs falling from about $2.50 to $0.50 after moving to Opus 5.5.
What people are saying
The thread largely accepts the core economic shift: generative AI can now turn a request and a query into a polished HTML dashboard cheaply enough that paying six figures for a visualization layer looks increasingly optional.
What readers liked
- AI-assisted dashboard generation makes bespoke HTML, SVG, and D3 visualizations practical without maintaining scarce front-end visualization…
- Reported Looker migrations suggest a part-time engineer plus Claude or Codex can validate a cheaper, Git-managed dashboard replacement surprisingly…
Concerns raised
- A model can turn a leading prompt and messy warehouse data into a convincing but false chart, making misinformation faster to produce and harder to…
- Semantic definitions remain the hard problem: giant, inconsistent schemas cannot simply be handed to an LLM and expected to yield reliable metrics.
Read full discussion analysis
Source: Hacker News · Based on 15 sampled comments · Analyzed
Discussion overview
The thread largely accepts the core economic shift: generative AI can now turn a request and a query into a polished HTML dashboard cheaply enough that paying six figures for a visualization layer looks increasingly optional. Several practitioners report real migrations from Looker or other SaaS stacks, using Claude or Codex to recreate dashboards in weeks or months and reducing a formerly expensive product to a version-controlled web app. AI-assisted D3 and unrestricted HTML/SVG are especially appealing because they remove the specialist front-end skills once needed for custom visual work.
But commenters repeatedly argue that charts were never the difficult part of business intelligence. The real work is deciding which metrics mean what, modeling messy source data, enforcing permissions, and preventing subtly wrong dashboards from becoming organizational fact. A language model can rapidly produce a persuasive visualization from an ambiguous prompt, which makes weak data governance more dangerous rather than less. Readers also push back on savings claims that count token costs but omit engineers’ time to build, validate, and maintain replacements. The emerging consensus is not that AI eliminates BI, but that it commoditizes the presentation layer while raising the value of semantic models, ETL discipline, provenance, guardrails, and analysts who can certify the numbers.
Points of disagreement
The central fight is whether AI-generated HTML can genuinely replace an enterprise BI product or merely replace its least valuable surface layer. Optimists point to successful Looker exits, dramatic licensing savings, and the freedom for anyone to request a tailored dashboard; skeptics argue that governed definitions, access control, data quality, and trustworthy metrics—not chart rendering—are what companies were paying for.
What readers liked
- AI-assisted dashboard generation makes bespoke HTML, SVG, and D3 visualizations practical without maintaining scarce front-end visualization expertise.
- Reported Looker migrations suggest a part-time engineer plus Claude or Codex can validate a cheaper, Git-managed dashboard replacement surprisingly quickly.
- Cheap natural-language generation lets teams prototype narrowly tailored data tools instead of forcing every question through a vendor’s fixed dashboard interface.
- When ETL and data modeling are already disciplined, agentic BI can spread usable operational insight beyond a small analyst team.
Concerns raised
- A model can turn a leading prompt and messy warehouse data into a convincing but false chart, making misinformation faster to produce and harder to spot.
- Semantic definitions remain the hard problem: giant, inconsistent schemas cannot simply be handed to an LLM and expected to yield reliable metrics.
- Token-price comparisons understate the cost of AI-built BI because someone still has to validate calculations, maintain the system, and own failures.
- Enterprise permissions and data sensitivity are not decorative features; an AI dashboard layer needs strong guardrails before broad self-service access is safe.
Keep reading
Sources and details
david.lequin.dev · Published
Can we replace a $150k Looker with a bucket of HTML files?
Read report →
Community analysis available for 1 of 1 linked discussions.
Additional discussion searches are still incomplete.
Coverage details and dates
Latest available community analysis:
Summaries: 1 current · 0 awaiting update · 0 not verified
Latest summary update:
Hacker News · david.lequin.dev