Cognition Labs

• No content whatsoever addressing Indigenous knowledge, data sovereignty, or relational epistemologies.
• The product frame ('autonomous software engineer') treats code as a culturally neutral, universal artifact.

• No acknowledgment of CARE Principles for Indigenous data governance.
• No consideration that codebases and training corpora ingest community-authored work without consent or attribution.
• No recognition of non-textual, oral, or relational ways of knowing as relevant to 'engineering'.

Total absence. Cognition's domain (code automation) is framed as universal and acultural, and Indigenous data sovereignty is entirely outside its stated concerns. Floor score.

• Some technical-historical humility in blog posts that share what did not work (e.g. 'Don't Build Multi-Agents', 'What We Learned Building Cloud Agents').
• Legal/compliance infrastructure (DPAs, terms) implicitly gestures at a regulatory inheritance.

• No engagement with colonial or extractive data legacies underlying code-trained models.
• No discussion of the geopolitical economy of compute (GPU access, supply chains) or the labor history of software production.
• Narrative is presentist: a clean origin story of 'expanding human capacity' with no contingency.

The only historical reflexivity is engineering postmortems. No structural-historical context about how the technology, data, or labor came to be. Near-floor.

• Product is implicitly English- and Western-software-convention-centric.
• No multilingual surface beyond the operating language of code.

• No multilingual support claims or localization commitments.
• No recognition that programming paradigms and 'good code' norms are culturally situated.
• No consultation with cultural scholars or non-Western engineering communities.

Code is presented as a universal logic with a single notion of quality ('good code'). No plurality of reasoning patterns or cultures is acknowledged. Low.

• Measurement orientation: 'AI Productivity Guarantee' ties claims to delivered engineering value.
• Candid technical postmortems disclose limitations rather than only successes.
• Devin Review introduced explicitly because generation quality is a real problem.

• Closed, proprietary API model with no open weights for independent verification.
• No independent third-party audits of training data, bias, or security claims cited.
• Productivity-guarantee framing is a commercial mechanism, not a published replication protocol; benchmarks/methodology not surfaced on these pages.

Above floor because of genuine measurement culture and honest limitation-sharing in engineering posts. Capped low-mid by closedness, absence of open weights, and no independent/third-party verification.

• Language is efficiency- and capacity-oriented ('exponential collaborator', 'expand human capacity').
• Craft language around code quality ('good code') gestures faintly at taste.

• No space for affective, intuitive, or ambiguous dimensions of making.
• No recognition of emotional labor or the felt experience of being automated alongside.
• Modes of attention beyond throughput/productivity are absent.

The register is engineering-productivist. 'Good code' is the closest nod to craft, but feeling, ambiguity, and emotional labor are absent. Low.

• Explicit anti-replacement framing: technology should expand capacity 'not by replacing meaningful work'.
• Engages displacement adjacent to its core claim by positioning Devin as collaborator rather than substitute.

• The replacement disclaimer sits in tension with marketing an 'autonomous software engineer' that 'ships production code independently' — contradiction left unexamined.
• No environmental/compute cost disclosure.
• No democratic or participatory governance of agentic systems; futures are shaped by the firm and its enterprise customers.

Credit for explicitly naming labor displacement and taking an anti-replacement stance. Held low by the unaddressed contradiction with 'autonomous'/'independently', no environmental disclosure, and no inclusive deliberation.

• Devin Open Source Initiative offers free access for maintainers — a community-facing, non-commercial gesture.
• Open-source maintainers identified as a constituency worth supporting.

• No participatory design with Global South developers.
• No disability/accessibility commitments.
• No labor-representative engagement or compensated feedback channels; 'community' = OSS maintainers and enterprise users, the already-connected.

One concrete inclusion gesture (free maintainer access) lifts it off the floor, but the table is otherwise enterprise- and elite-developer-shaped, with no marginalised constituency named. Low.

• Genuine willingness to contradict industry consensus: 'Don't Build Multi-Agents' inverts a popular trend.
• Self-critical framing that generation is easy and review is the real bottleneck punctures AI-coding hype.

• The central paradox — selling an 'autonomous engineer' while insisting work is not being replaced — is never named or played with.
• Irony is deployed at competitors'/the field's expense, not turned on Cognition's own seriousness.
• No structural space where the official 'expand human capacity' narrative is tested by its opposite.

Above floor for real contrarian, consensus-puncturing posts that show appetite for inversion. Capped low because the inversion is never aimed inward at the firm's own foundational contradiction.