About

What is Smol Gardens?

A Civic Builder's Guide to Accountable Tech

Origins

Smol Gardens began as a concept presented by the Femmecubator team at BetaNYC's UnSchool of Data 2026 conference on March 2026, emerging from their Open Civic Tech initiative. The second workshop was presented at Makeshift 2026: Accountable Tech by Design event during NYC Design Week.

The project's name draws inspiration from Anil Dash's 2012 talk "The Web We Lost," which critiqued the internet's evolution into isolated, walled gardens. Today, as AI reshapes the technological landscape, Smol Gardens invites civic technologists to challenge LLM platform usage and explore alternative AI tools in civic tech tool building.

Problem Statement

Designers, builders, and civic technologists are facing a core problem: how to develop civic tech responsibly in the age of AI without creating redundant, unmaintainable, or harmful solutions.

Is it possible to create meaningful work using LLM models? When a new wave of vibecoders (builders with no programming skills) are shipping websites in less than a day, there's a clear gap in accountability that needs to be addressed.

This creates several cascading challenges:

  1. Lack of frameworks for responsible AI-enabled design — there's no clear, accessible methodology that allows practitioners to balance speed with responsibility while building civic tech.
  2. Lack of transparency and unaccountability — current AI development practices often lack transparency about tradeoffs, making it difficult for communities to understand how AI systems will impact them or to offer meaningful critique.
  3. Scale mismatch — large language models (LLMs) are presented as the default solution, but they may be overkill for many civic tech applications and come with disproportionate environmental and resource costs.
  4. Weakened community agency — without continuous accountability mechanisms and openness to critique, affected communities lose the ability to shape and govern the AI systems being built for them.
  5. No clear alternative path — practitioners who want to build differently lack both a practical toolkit and evidence that smaller, more efficient AI models can deliver meaningful civic tech solutions.

Our Proposal

Smol Gardens establishes a groundwork for iterative study through three phases:

  1. Impact Framework

    Create the impact assessment tracking tool. We're developing a framework that measures human, systems, and environmental impact on AI-enabled web-development.

  2. Design Workshops

    Invite civic technologists document actual AI usage through a "diary mission," using the Impact assessment tool as guide.

  3. SLM Platform

    Use small language models (SLMs) to build a curated repository of civic tech tools hosted by the Accountable Tech working group and will be made available open source for future builders.

Who Is This For?

This workshop is for civic technologists who are already using AI tools in their workflow. Participants will be invited to document and audit their work using an impact assessment framework. The study aims to educate and push for alternatives to using LLM models, with an end goal to incorporate Accountable tech principles in a civic builder's workflow.

Framework

Guiding Principles

What is Responsible Computing?

Responsible computing recognizes that technology is NOT neutral. Every algorithm, dataset, system design, and infrastructure choice reflects human power structures, cultural values, and societal assumptions. Our role as technologists is to:

  • Identify where these choices are embedded
  • Determine where we can intervene responsibly and effectively
  • Make visible the human choices that affect real people

Four Core Principles

1. Technology as Situated

Algorithms and datasets are not objective. They encode historical dynamics, cultural norms, institutional practices, and human biases—the values and assumptions of the people who created them.

  • Understanding the context of creation is essential
  • Recognizing what values technology carries forward is critical

2. Stakeholder Impact Analysis

Technical projects affect different groups unequally. Marginalized and vulnerable populations often experience disproportionate harm. Critical questions to ask:

  • Who benefits from this system? Which groups gain advantages or privileges?
  • Who bears the risks? Who could be harmed or excluded?
  • Whose interests are not represented? Which voices are missing?
  • Which communities are most vulnerable to harm? Who is most at risk?

3. Power and Value Choices

Every technical decision involves value judgments. Trace these choices through hardware architecture, software implementation, and data science models. Make choices visible and contestable.

  • Who has power in system design? Who makes final decisions?
  • What economic incentives guide implementation? What drives prioritization?
  • How are trade-offs decided? When features conflict, whose needs win?
  • Whose preferences are prioritized? Whose are sacrificed?

4. Multidisciplinary Evaluation

Ethical dilemmas cannot be solved through quantitative metrics alone. Bring qualitative tools from STS (Science, Technology, and Society) alongside technical analysis. Avoid isolated decision-making and engage:

  • Stakeholders affected by the system
  • Historians and social scientists
  • Affected communities
  • People with lived experience of the problem you're solving

The HCE Toolkit Concepts

The Human Contexts & Ethics (HCE) Toolkit provides analytical concepts to systematically examine where human choices are embedded in technology:

  • Power: Who controls decisions? How do systems reinforce or reshape authority?
  • Representation: What assumptions are embedded in data and models? How do technical representations shape identity?
  • Sociotechnical Systems: How do humans and technology interdependently shape outcomes? Where do risks concentrate?
  • Identity & Positionality: How are gender, race, class, disability, and other identities embedded in systems? Whose perspectives are missing?
  • Labor & Materiality: What human effort sustains the system? What are the environmental and physical consequences?
  • Co-production: How do technology and society shape each other? What historical assumptions persist?
  • Expertise & Institutions: Who is credentialed to decide? What institutions legitimize certain knowledge over others?
  • Sociotechnical Imaginaries: What collective visions of the future are embedded in system design?

Before Launch: Critical Questions

Use this checklist to systematically apply responsible computing principles:

  • Stakeholder Analysis: Who are all the stakeholders affected by this system, including those not yet harmed?
  • Power & Values: Where are power asymmetries embedded in design and decision-making?
  • Historical Context: What historical or cultural assumptions are we encoding into this technology?
  • Benefits & Harms: Who benefits? Who bears the costs? Are these distributed equitably?
  • Representation & Identity: How does this system represent people? Could it be used to discriminate?
  • Expertise & Legitimacy: Whose expertise was included? Whose was excluded?
  • Intervention Points: Where can we mitigate harms or increase equity?

Key Takeaway

Responsible computing is not a compliance checklist or a box to check before launch. It is a disciplined practice of making visible the human choices embedded in technology, analyzing their consequences for different groups, and determining where you can intervene to build systems that are more equitable, transparent, and accountable.

Framework

Current Challenges

I. Current State of the System

This work addresses a fundamental question: Accountability in Tech, specifically with the process of vibecoding or rapid prototyping with AI. Vibecoding offers accelerated development and efficiency gains, yet it presents a complex tradeoff. Can we harness its benefits while mitigating its harms?

Key Question

Can we build responsibly in the Age of AI?

Policy Lag

Ethical AI policy and legislation are trailing far behind the speed of deployment. The technology is already being misused — malvertising via vibecoded sites, unauthorized use of writers' and designers' work, and with new products promising to be a job killer for designers, diminishing the value of human creative labor.

Large orgs see AI-powered development as a cost-cutting alternative, one that could replace one thousand interns. Yet the reality is more complicated. Data centers are proliferating to support a trillion-dollar LLM industry, creating staggering environmental costs. When we apply second-order thinking, deeper issues emerge: the process introduces redundancy, creates maintenance burdens, raises data sovereignty concerns, and opens vectors for malware.

Environmental Impact

The computational demands of training large language models carry a significant environmental burden that is often overlooked. Training a single common AI model can emit more than 626,000 pounds of CO2 equivalent, approximately five times the lifetime emissions of an average car, including manufacturing.

This environmental cost raises critical questions: if we're building AI solutions to serve communities responsibly, can we justify deploying models with such substantial carbon footprints for applications that smaller, more efficient models might adequately serve? The environmental impact of AI development is not merely a sustainability concern; it represents a failure of accountability to future generations and disproportionately affects communities already vulnerable to climate impacts, making responsible model selection a matter of environmental justice in civic tech.

Source: UNESCO's Recommendation on the Ethics of Artificial Intelligence

II. Workshop Results: BETANYC Unschool of Data Session

We brainstormed with 15 attendees and captured themes from their feedback on this issue:

Discussion Points

  1. Compliance & Accessibility Gaps — 508 compliance, security best practices missing
  2. Maintenance & Scalability Issues — Code works initially, fails to scale
  3. Convergence Toward Average — All LLM solutions look the same, everything is a result of statistical average / vanilla quality content / slop
  4. Skill & Community Erosion — Automation replaces human collaboration and community building
  5. Environmental & Health Consequences — Real harm to vulnerable communities (Memphis data center example)
  6. Ownership & Accountability Gaps — No clear governance or long-term responsibility
  7. Black Box Training Data — Unknown sources, no attribution, no consent
  8. Bias & Marginalized Voices — Solutions amplify privileged perspectives, silence marginalized groups
  9. Transparency & Accountability — Attribution, consent, open knowledge commons
  10. Resource Efficiency — Prevent redundancy, reuse existing solutions

The original discussion notes are posted on this document.

Framework

Data Index Checklist

Overview

This framework is recommended for civic tech builders to evaluate their workflow using a the IAT (Impact Assessment Tool) scorecard that contains 15 questions. The scorecard uses three core pillar categories: Human, Environment and Systems Impact.

Example Evaluation Checklist

Below is how a project might be assessed across key criteria:

Meets Core Purpose — Is the solution mission-aligned?
Uplift of Labor & Skills — Do participants gain new skills?
Reusable Work for Collective — Is work documented and shareable?
Cognitive Sovereignty — Did builders make deliberate tool choices?
Environmental Impact — Measured and documented?

Did participants gain capability, ownership, and transferable knowledge through this process?

Core Purpose

Is the solution still aligned with the civic problem it was designed to solve — or has it drifted toward efficiency for its own sake?

Uplift of Labor & Skilling

Did building this tool expand participants' technical, collaborative, or domain capabilities?

Cognitive Debt

Did the team accumulate cognitive debt — outsourcing reasoning to AI in ways that leave no one able to explain or maintain decisions?

Creative Agency & Decision Making

Did participants make deliberate choices about which tools to use, or did they default to AI suggestions?

Harm, Exclusion & Manipulation

Could this tool harm, exclude, or manipulate any member of the community it serves?

Did we build something communities can own, maintain, and trust without depending on us?

Reusability & Redundancy

Is the methodology documented and shareable so other builders can build on it without starting from scratch?

Quality & Trust

Were real-world harms understood and documented before the tool went live?

Corporate Dependency

Does the community retain meaningful control without depending on platforms they don't own?

Systemic Fragility

Is there a maintenance plan sufficient for the tool to survive without its original builders?

Transparency & Explainability

Can end users understand how the tool works and why it produces the outputs it does?

Safety & Privacy

Have the safety risks and privacy implications for the community been identified, documented, and mitigated?

Did we take responsibility for the environmental cost of the AI we used?

Energy Consumption & Carbon Footprint

Did we track and document the carbon emissions generated by our inference calls?

Inference Calls & Model Size

Did we deliberately choose the smallest model capable of the task and log the number of inference calls made?

Environmental & Resource Cost

Did we account for the full resource cost — energy, carbon, and water — and publish our findings for others to replicate?

Community Burden

Does the energy infrastructure powering this AI displace resources or burden communities — local or future — who had no say in that choice?

Design Challenge

30-Day Challenge

Overview

The Accountable AI 30-Day Challenge is an intensive, collaborative experiment bringing together civic tech builders, designers, and community members to build meaningful tools using ethical AI practices.

Who Can Participate

  • Civic Tech Builders: Developers, designers, and technologists working in the civic space
  • Domain Experts: Education, democracy, and social change specialists
  • Community Members: People directly affected by civic technology decisions
  • Crit Reviewers: Senior practitioners providing feedback and guidance

What You'll Build

Choose one of two civic domains and develop a real project that addresses a community need:

  • Tools for Education: Tools, frameworks, or resources that improve learning
  • Tools for Democracy: Civic engagement platforms, information systems, or participation tools

Key Commitments

  • Use small language models (Qwen, SmolLM, Haiku 4.5) instead of large language models
  • Document your process daily through video diaries or written logs
  • Participate in weekly group check-ins and feedback sessions
  • Publish your work as open-source at the end of the challenge
  • Complete the impact assessment tool to evaluate your project
Design Challenge

Project Plan

30-Day Structure

Phase Days Focus
Discovery 1–5 Problem identification, community interviews, POV statement writing
Ideation 6–12 Concept sketching, tool selection, architecture planning
Build 13–25 Prototype development, user testing, iteration cycles
Polish & Document 26–30 Finalization, environmental audit, report writing, presentation prep

Daily Cadence

  • Individual Work: 2–3 hours per day on your project
  • Documentation: 30 minutes of journaling or video diary entries
  • Weekly Check-in: 1 hour group meeting for peer feedback and updates
  • Crit Sessions: Days 28–30 final presentations and structured feedback

Team Roles & Commitments

Role Responsibilities Commitment
Lead Facilitator Overall coordination, orientation facilitation, final report authorship Full-time, 30 days
Participants (5–10) Project selection, building, daily documentation, evaluation, design crit ~2–3 hrs/day
Crit Reviewers (3–5) Attend final design critique, provide written feedback Days 28–30 only
Open-Source Community Challenge first drafts, contribute to the Smol Gardens repo. Ongoing / voluntary
Responsible Computing Principles

Guiding Principles

What is Responsible Computing?

Responsible computing recognizes that technology is NOT neutral. Every algorithm, dataset, system design, and infrastructure choice reflects human power structures, cultural values, and societal assumptions. Our role as technologists is to:

  • Identify where these choices are embedded
  • Determine where we can intervene responsibly and effectively
  • Make visible the human choices that affect real people

Four Core Principles

1. Technology as Situated

Algorithms and datasets are not objective. They encode historical dynamics, cultural norms, institutional practices, and human biases—the values and assumptions of the people who created them.

  • Understanding the context of creation is essential
  • Recognizing what values technology carries forward is critical

2. Stakeholder Impact Analysis

Technical projects affect different groups unequally. Marginalized and vulnerable populations often experience disproportionate harm. Critical questions to ask:

  • Who benefits from this system? Which groups gain advantages or privileges?
  • Who bears the risks? Who could be harmed or excluded?
  • Whose interests are not represented? Which voices are missing?
  • Which communities are most vulnerable to harm? Who is most at risk?

3. Power and Value Choices

Every technical decision involves value judgments. Trace these choices through hardware architecture, software implementation, and data science models. Make choices visible and contestable.

  • Who has power in system design? Who makes final decisions?
  • What economic incentives guide implementation? What drives prioritization?
  • How are trade-offs decided? When features conflict, whose needs win?
  • Whose preferences are prioritized? Whose are sacrificed?

4. Multidisciplinary Evaluation

Ethical dilemmas cannot be solved through quantitative metrics alone. Bring qualitative tools from STS (Science, Technology, and Society) alongside technical analysis. Avoid isolated decision-making and engage:

  • Stakeholders affected by the system
  • Historians and social scientists
  • Affected communities
  • People with lived experience of the problem you're solving

The HCE Toolkit Concepts

The Human Contexts & Ethics (HCE) Toolkit provides analytical concepts to systematically examine where human choices are embedded in technology:

  • Power: Who controls decisions? How do systems reinforce or reshape authority?
  • Representation: What assumptions are embedded in data and models? How do technical representations shape identity?
  • Sociotechnical Systems: How do humans and technology interdependently shape outcomes? Where do risks concentrate?
  • Identity & Positionality: How are gender, race, class, disability, and other identities embedded in systems? Whose perspectives are missing?
  • Labor & Materiality: What human effort sustains the system? What are the environmental and physical consequences?
  • Co-production: How do technology and society shape each other? What historical assumptions persist?
  • Expertise & Institutions: Who is credentialed to decide? What institutions legitimize certain knowledge over others?
  • Sociotechnical Imaginaries: What collective visions of the future are embedded in system design?

Before Launch: Critical Questions

Use this checklist to systematically apply responsible computing principles:

  • Stakeholder Analysis: Who are all the stakeholders affected by this system, including those not yet harmed?
  • Power & Values: Where are power asymmetries embedded in design and decision-making?
  • Historical Context: What historical or cultural assumptions are we encoding into this technology?
  • Benefits & Harms: Who benefits? Who bears the costs? Are these distributed equitably?
  • Representation & Identity: How does this system represent people? Could it be used to discriminate?
  • Expertise & Legitimacy: Whose expertise was included? Whose was excluded?
  • Intervention Points: Where can we mitigate harms or increase equity?

Key Takeaway

Responsible computing is not a compliance checklist or a box to check before launch. It is a disciplined practice of making visible the human choices embedded in technology, analyzing their consequences for different groups, and determining where you can intervene to build systems that are more equitable, transparent, and accountable.