Before You Commit to Technology, Ask What It Will Make of You
A culturally minded framework for choosing tools, platforms, partners, and technical systems without surrendering taste, independence, or the freedom to change course.
Marek DvořákSenior product reviewerFirst published 9/19/2026 · monitored for updates; the next revision publishes a new version and appears here. Reader corrections are reviewed and folded into future versions.
Summary
Technology rarely arrives as a neutral purchase. A platform changes how work is seen; a tool influences what gets made; an infrastructure choice quietly determines who owns the archive and how expensive departure becomes. For founders, artists, architects, designers, and cultural institutions, the decisive question is therefore not merely whether a technology works, but what habits, dependencies, aesthetics, and futures it installs. Before committing, examine the promise, the exit, the hidden labor, and the kind of culture the system rewards.
Key takeaways
- Define the human or cultural need before comparing features.
- Treat reversibility as a design quality: ask how data, files, audiences, and workflows can leave.
- Calculate total commitment, not only subscription price—including training, integration, migration, energy, and attention.
- Interrogate defaults: they encode assumptions about authorship, speed, visibility, privacy, and taste.
- Request evidence from a context resembling yours; a Silicon Valley case study may not translate to an Icelandic studio or museum.
- Distinguish a useful prototype from infrastructure that must remain dependable for years.
- Name what must never be automated, especially judgment, consent, attribution, and culturally sensitive interpretation.
- Set a review date before signing, so continuation becomes a deliberate decision rather than inertia.
Deep dive
Begin with the life around the tool
The weakest technology decisions begin with a product category: ‘We need AI,’ ‘We need an app,’ or ‘We should move to the cloud.’ Begin instead with a scene. Who is doing what, in which place, under what constraints—and what would become meaningfully better? A Reykjavík design studio handling large visual files has different needs from a rural craft workshop documenting techniques, a festival managing temporary staff, or a museum stewarding sensitive collections. Write the desired change in ordinary language and include a non-technical alternative. If clearer governance, fewer meetings, or a better spreadsheet solves the problem, new software may only add ceremony. Also ask who requested the technology and who will carry it. The buyer often receives the presentation; the worker inherits the interface.
Read the business model as part of the interface
Every tool has two designs: the product people use and the economic machine sustaining it. Ask who pays, what investors expect, and whether growth depends on advertising, surveillance, transaction fees, cloud consumption, or capturing a market. A free platform may be financed by behavioral data; a low introductory rate may presume future switching costs; an AI service may subsidize adoption before increasing prices. Examine the vendor’s funding, profitability, acquisition history, terms, and dependence on another provider’s infrastructure. For a small organization, resilience matters more than a dazzling roadmap. Will the company answer support requests? Can it survive a downturn? If acquired, can the new owner change terms, discontinue a feature, or train models on uploaded material? Technology procurement is partly counterparty analysis.
Inspect the exit before admiring the entrance
A good commitment includes a dignified way out. Request a sample export before signing: not a promise of portability, but the actual formats, metadata, relationships, and media returned. Open files can still be practically unusable if captions, permissions, version histories, or links disappear. Ask how deletion is verified, how long backups persist, whether APIs cost extra, and who pays for migration. For creative work, clarify ownership of source files, fonts, prompts, outputs, training data, and derivative material. For audience platforms, remember that followers are not necessarily portable relationships. Email addresses held with informed consent are different from reach rented through an algorithm. Reversibility is not pessimism; it preserves negotiating power and encourages experimentation.
Audit the aesthetics and the invisible labor
Tools teach a style. Templates normalize certain proportions; recommendation systems favor recognizable patterns; generative systems tend toward the statistically available. Ask whether the technology expands a distinctive practice or compresses it into platform-native sameness. Icelandic cultural work can be especially vulnerable when globally trained systems flatten language, place names, folklore, craft knowledge, or landscape into generic Nordic imagery. Test edge cases in Icelandic and English, including diacritics, accessibility, attribution, and culturally specific terms. Then count the labor hidden behind apparent convenience: cleaning data, correcting outputs, moderating abuse, securing accounts, writing alt text, gaining consent, and supporting colleagues. Automation often relocates work rather than removes it—and may relocate it to the least powerful person in the room.
Measure consequences at the right scale
A pilot can look efficient while exporting costs elsewhere. Evaluate security, privacy, accessibility, environmental demand, and social effects alongside speed. The International Energy Agency estimated that data centres, AI, and cryptocurrency used nearly 460 terawatt-hours of electricity globally in 2022 and projected demand could exceed 1,000 TWh in 2026. That does not make every digital project irresponsible; it makes scale, model choice, hosting region, device lifespan, and frequency of use material design decisions. In Iceland, abundant renewable electricity does not erase questions about land, grid capacity, hardware, water, or opportunity cost. Ask what happens if usage grows tenfold, if sensitive data leaks, or if a person cannot use the system. Small tests should measure harms as seriously as engagement.
Turn enthusiasm into a bounded experiment
Commit in stages. Define a 30- to 90-day pilot with named users, a spending ceiling, success and failure criteria, prohibited uses, and a stop date. Track quality, time saved after correction, adoption, incidents, accessibility, and the emotional texture of the work—not just output volume. Assign an accountable owner and an independent skeptic. For consequential systems, document decisions in an impact assessment and maintain a manual fallback. At review, ask whether the tool strengthened capability or merely increased dependency. The best technology leaves an organization more articulate about its values, more capable of acting, and freer to choose what comes next.
- 1985The Free Software Foundation formalized a movement linking software use to freedoms to inspect, modify, and share code.
- 1989Tim Berners-Lee proposed the World Wide Web at CERN, building on open standards rather than a single proprietary network.
- 2001The Agile Manifesto prioritized people, working software, collaboration, and adaptation over rigid process.
- 2006Amazon Web Services launched S3 and EC2, accelerating the shift from owned servers to rented cloud infrastructure.
- 2016The European Union adopted the General Data Protection Regulation; enforcement began on 25 May 2018.
- 2021UNESCO adopted its Recommendation on the Ethics of Artificial Intelligence, addressing rights, culture, environment, and governance.
- 2022OpenAI released ChatGPT publicly, making generative AI procurement and authorship questions immediate for creative organizations.
- 2024The EU AI Act entered into force on 1 August, beginning phased obligations based on system risk.
Glossary
- Reversibility
- The practical ability to abandon a system without losing essential data, capability, relationships, or unreasonable sums of money.
- Vendor lock-in
- Dependence created when switching providers is technically difficult, contractually restricted, or prohibitively expensive.
- Interoperability
- The capacity of different systems to exchange information and use it meaningfully through shared formats or standards.
- Total cost of ownership
- Purchase price plus implementation, training, support, integration, compliance, maintenance, migration, and eventual retirement.
- Algorithmic opacity
- A condition in which users cannot adequately understand how a system reaches decisions or ranks content.
- Data provenance
- A record of where data originated, how it was gathered, and what transformations or permissions apply to it.
- Human-in-the-loop
- A workflow that reserves defined review, correction, or approval decisions for people rather than full automation.
- Technical debt
- Future work and risk created by expedient technical choices that later require repair, replacement, or continued compromise.
- Open standard
- A publicly documented specification that multiple parties can implement, reducing dependence on one supplier.
FAQs
What is the first question to ask before buying technology?+
Ask what specific human condition should improve and for whom. If the answer is only ‘innovation,’ ‘efficiency,’ or ‘visibility,’ the problem is not yet sufficiently defined.
How can a small creative business assess a vendor?+
Review ownership, funding, pricing history, security documentation, export options, support terms, and credible customer references. Ask what occurs if the company is acquired, closes, or retires the feature you need.
When is a pilot meaningful rather than performative?+
A meaningful pilot has representative users, real material, measurable criteria, prohibited uses, and a predetermined end date. It must be possible for the result to be ‘do not adopt.’
Should artists avoid generative AI?+
Not categorically; the relevant questions concern consent, provenance, attribution, aesthetic purpose, and contractual rights. Use should strengthen rather than obscure the artist’s agency, and confidential or culturally sensitive material should not be uploaded casually.
How should an Icelandic organization test language quality?+
Test real Icelandic text, inflection, names, diacritics, speech, and specialist vocabulary with fluent reviewers. Do not infer competence from an English demonstration or from a vendor’s broad claim of multilingual support.
Is open-source software always safer from lock-in?+
No. Source access can improve inspectability and switching power, but specialist skills, hosting architecture, proprietary extensions, or a weak maintainer community can create other dependencies.
Which contract clauses matter most?+
Prioritize data ownership, permitted model training, confidentiality, breach notification, service levels, price changes, subcontractors, termination, deletion, export format, and liability. For important systems, obtain legal and security review appropriate to the risk.
How often should a technology commitment be reviewed?+
Review at least annually and sooner after material price, ownership, policy, security, or product changes. High-risk or rapidly evolving AI systems may require quarterly monitoring.
Risks
- A convenient proprietary workflow can become an expensive archive: files technically export, but essential metadata, permissions, links, or edit histories do not.
- Creative teams may mistake faster production for better work, increasing generic output while eroding research, craft, authorship, and trust.
- Sensitive client, cultural, or personal material may be retained by vendors, routed through undisclosed subprocessors, or used under changing terms.
- Automation can centralize benefits while shifting correction, moderation, and accountability onto junior staff, contractors, or affected communities.
- A pilot built on unusually attentive users and vendor support may conceal the training, security, accessibility, and environmental costs of organization-wide deployment.
Opportunities
- Build ‘exit-ready’ creative operations around open formats, clean metadata, local backups, and documented workflows; resilience becomes both a business advantage and a cultural preservation practice.
- Create Icelandic-language evaluation datasets and human review services for tourism, public culture, design, archives, and media—areas where global benchmarks remain inadequate.
- Develop procurement studios that combine service design, contract review, accessibility, carbon literacy, and aesthetic direction for small cultural organizations.
- Use narrowly scoped automation to remove administrative repetition while reserving interpretation, selection, storytelling, and relationship-building for people.
- Turn transparent technology choices into a trust signal: publish provenance, AI-use disclosures, data policies, and the reasons certain processes remain deliberately human.
Sources & references
- NIST AI Risk Management Framework 1.0
- OECD Principles on Artificial Intelligence
- UNESCO Recommendation on the Ethics of Artificial Intelligence
- Regulation (EU) 2016/679 — General Data Protection Regulation
- Regulation (EU) 2024/1689 — Artificial Intelligence Act
- IEA Electricity 2024: Analysis and Forecast to 2026
- The Green Software Foundation: Software Carbon Intensity Specification
- Web Content Accessibility Guidelines (WCAG) 2.2
| Reversible pilot | Managed platform | Custom or self-hosted system | |
|---|---|---|---|
| Initial commitment | Low; 30–90 days, capped users and budget | Medium; contract, onboarding, and integrations | High; architecture, staffing, maintenance, and governance |
| Speed to useful evidence | Fast, if tested with real work | Usually fast after configuration | Slowest; discovery and implementation precede evidence |
| Control | Limited but intentionally temporary | Moderate; bounded by vendor roadmap and terms | High over data, features, hosting, and release timing |
| Exit difficulty | Low when exports and deletion are tested first | Medium to high as integrations and archives accumulate | Variable; code is controllable, but skills and infrastructure may lock you in |
| Best fit | Uncertain needs, emerging AI, one-off experiments | Stable common workflows with limited technical staff | Distinctive capability, sensitive data, or long-lived cultural assets |
| Primary question | What evidence would justify continuing? | Can we tolerate this vendor’s future choices? | Can we fund stewardship after launch? |
As electric vehicles loosen the cabin from the mechanical logic of combustion, a new design territory emerges: part salon, part screen, part sound studio—and potentially the most consequential room in motion.
A durable guide to reading the technological frontier—where artificial intelligence, spatial computing, robotics, biotechnology, and climate systems become new materials for culture, products, and companies.
When language becomes an interface, ordinary words can behave like executable instructions. Here is how prompt injection works, why it resists simple fixes, and where thoughtful builders can create safer, more legible AI products.
A field guide to the shift from cloud-dependent intelligence toward private, responsive, context-aware computing at the edge—and the products it makes possible.
From our own rounds
Measured on The Curator, from real sessions people played on this site — not a third-party dataset.
- Rounds played here
- 143
- Questions per round
- 1.7