01
Published
After the pyramid: what changes at work when AI performs more
AI amplifies execution, but purpose, learning, judgment and accountability depend on how we redesign work.
Automatic publishing · America/Sao_Paulo
What has already come out.
What comes next.
The agenda is generated by the same contract that releases the index, feed and sitemap. Times indicate eligibility; deployment occurs on the first successful automatic cycle after the time.
02
Published
The model is not the system: why powerful AI still needs a harness
Models change quickly; reliable results depend on context, tools, policies, verification, evidence and learning.
03
Published
Accountable human, executing AI: the contract that organizes the squad
AI can perform more work without receiving accountability: clear roles, boundaries, verifiers, and evidence preserve accountability.
04
Published
Why Trustyu created FORGE: from a founder's memory to a verifiable system
JARVIS's journey to FORGE and the six changes that transformed AI memory, decisions, and execution into a verifiable system.
05
Published
AI amplifies the business you already have
AI does not transform an organization alone: it enhances clarity, flow and learning — or accelerates ambiguities, queues and rework.
06
Published
Engineering harness in practice: six plans to transform intention into evidence
A technical guide for organizing intent, policy, execution, verification, evidence, and feedback around AI agents.
07
Published
From Pilot to Value: Measure Accepted Change, Not Produced Tokens
A practical contract to measure AI pilots by accepted outcomes, quality, flow, total cost, and risk — not by tokens or artifacts.
08
Published
Spec-driven without self-deception: coverage, contract, eval and OpenAPI are not the same proof
How to separate specification, contract, test, coverage and eval so that each green supports only the decision it actually verified.
09
Published
Autonomy proportional to risk: where to delegate to AI and where to keep the human gate
A method for grading agent authority by effect, reversibility, sensitivity and uncertainty — without confusing capacity with permission.
10
Published
Evidence before adjective: how to defend claims about AI systems
A practical grammar to link claims to subject, property, mechanism, evidence, verifier, validity and limit of inference.
11
Published
The agent cannot be its own auditor
How to separate executor, checker, reviewer and approver to use AI in evaluation without creating circular authority or fictitious independence.
12
Published
From idea to first revenue: P1–P6 for AI-first founders
A six-phase path to turn building speed into evidence of a problem, product, operation, and a real exchange of value.
13
Published
AI lowers the cost of experimenting; it does not validate your hypothesis
How to use AI to prepare better experiments without promoting a synthetic persona, click, prototype, or opinion into market evidence.
14
Scheduled
From checklist to executable control: governance that actually changes the system
A practical contract to turn AI rules into properties, applicability, mechanisms, evidence, authority, and validity.
15
Scheduled
When AI fails, does the system learn or merely retry?
How to turn agent failures into regressions, controls, evals, and durable decisions instead of repeating retries across sessions.
16
Scheduled
From buying copilots to building an AI-first organization
How to turn individual gains from copilots into workflows, platforms, authority, evidence, and organizational learning.
17
Published
The model became cheaper; the system became expensive: the new AI cost chain
The technical AI cost chain, from inference through context, harness, CI, evals, operations, human work, failures, and capacity.
18
Scheduled
Long-running agents: context, handoffs, and progress across sessions
The continuity contract that preserves intent, state, decisions, evidence, and resumption when tasks span sessions.
19
Scheduled
Graph, loop, or hybrid? How to choose orchestration without handing control to the model
Criteria for choosing a chain, graph, loop, or hybrid with budgets, stop conditions, authority, and evidence defined by the system.
20
Published
Your new portfolio should prove how you think, verify, and learn
A professional case-study model that shows the problem, decisions, use of AI, verification, outcome, limits, and learning.
21
Scheduled
Junior professionals in the AI era: how to grow when execution became cheaper
A learning path for using AI as a tutor and amplifier without outsourcing understanding, verification, judgment, and accountability.
22
Published
Sandbox, capability and identity: task-based security in agentic systems
A security contract per task that separates workspace, sandbox, identity, capability, data, time, policy and evidence.
23
Published
Taste, responsibility and clarity: the human skills that have become more valuable
When producing becomes cheaper, clarity, taste, critical thinking, responsibility and learning raise the standard of work.
24
Published
Solo founder + AI squad: speed without outsourcing judgment
How a founder can orchestrate research, product, architecture and engineering with AI without outsourcing priority, risk or acceptance.
25
Published
The first product is not just a product: it is the factory for the second
The first product must also leave standards, contracts, pipelines and learnings that reduce the cost and risk of the next one.
26
Published
Context engineering: what deserves to enter, leave and survive the session
How to select, structure, retrieve, compress, externalize and expire context without confusing chat with durable memory.
27
Published
Evals before release: how to transform expectations into executable criteria
Tasks, graders, datasets, trials, baseline and gates make an agent's expectations observable before releasing a change.
28
Published
The right queue for AI: prioritize decisions and outcomes, not flashy automations
A method for prioritizing AI opportunities by outcome, frequency, friction, verifiability, risk and learning.
29
Published
Seniority in the age of agents: fewer ready-made answers, more standard of proof
Seniority appears in framing, judgment, verification, ownership, multiplication and limits — not in the volume produced.
30
Published
Design Partner is not a market: proximity to learn without inventing PMF
How to learn deeply from a Design Partner without turning intense collaboration into false proof of market or PMF.
31
Published
Fable, Sol, and Terra: what really changed for builders
A dated radar of new AI portfolios and a method for choosing by task, risk, cost per outcome, evals, and verifiability.
32
Published
Shadow builders: controls and evidence to operate AI-Native software
The RedAccess case converted into three model, discovery, negative tests and Evidence Gate for applications quickly created with AI.
33
Published
AI writes, AI reviews: independence needs to be demonstrated
The review study IA-for-IA converted into authority segregation, protected checks, human approval and release provenance.
34
Scheduled
A security prompt is not a control: how to prove properties in AI-generated software
The paper on security prompts turned into a technical contract for specification, protected verifiers, tests, authority, and an Evidence Packet.
35
Published
It passed the tests. Is it ready to operate?
A proposed protocol to distinguish functional correctness, behavior under load and recovery capability in services built with AI assistance.
36
Scheduled
AI code review: from the approval queue to verifiable supervision
How to turn architectural intent into checks, preserve approval authority, and require evidence for the version to be released.
37
Published
AI-Native Product Lead: the contract between intent, agents and evidence
A verifiable design to connect value, architecture, evals, risk, economy and human authority in products built with AI.
38
Published
Agent Control Standard: how to prove controls on runtime without calling warranty preview
A technical reading of the OWASP ACS preview v0.1, with integration architecture, minimum tests, limits of evidence and adoption criteria.
39
Published
Formal proof and verifiable outputs: when the verifier settles the question
An Evidence Brief on proof-carrying outputs, independent verifiers, and what the formalization of Fermat teaches agent systems.
40
Published
The repository is not trustworthy: controls for agents at the workspace–runtime boundary.
Evidence Brief on monotonic configuration, approval, confinement, environment, and network for agents operating on untrusted code.
41
Published
Localhost is not identity: controls for agent local APIs
Evidence Brief on identity, authorization, CORS, control planes, sandboxing, and receipts for local APIs used by AI agents.
42
Published
IT does not disappear: it becomes the platform for democratizing AI
Minimum architecture to transform distributed AI experiments into identified, governed and operable assets without recreating the IT queue.
43
Published
Three gates for production agents: boundary, stopping, and consensus
Evidence Brief on SSRF across redirects, stopping conditions for iterative repair, and the use of ensembles as a review queue rather than truth.
44
Published
FDE without improvisation: the harness that turns the customer mission into a verifiable product
The technical contract that connects mission, discovery, spec, build, evals, production, adoption, and reusable learning in an FDE's work.
45
Published
Reviewing code with AI requires adversarial knowledge and dynamic validation
Evidence Brief on BlueCodeAgent, adversarial knowledge, dynamic analysis, and the limits of using AI to review code security.
46
Published
Specifying before testing improves detection — but costs 38% more tokens
Evidence Brief on contract-driven testing, detection gains on 90 production bugs, and the additional cost of turning intent into evidence.
47
Published
Explainable AI in the software chain: detecting anomalies is not proving security
Evidence Brief on XAI-SCS, anomaly detection in npm and PyPI packages, actionable explanations, and the boundaries between alerting, investigation, and security.
48
Published
The test passed, but did the agent fix or rewrite?
Evidence Brief on the Precise Debugging Benchmark and the difference between passing tests, recovering the expected fix, and avoiding unnecessary changes.
49
Published
Auto-approve is not a neutral convenience
Evidence Brief on HarnessSecurity-Bench, coding-agent authority, and the trade-off among utility, human confirmation, and adversarial effects.
50
Scheduled
Parallel sub-agents do not guarantee faster delivery
Evidence Brief on dynamic concurrency in coding agents, coordination cost, integration failures, and conditions for useful parallelism.
51
Scheduled
The PoC passed. Did the patch follow maintainer intent?
Evidence Brief on the reliability of patch benchmarks, PoCs, regression, developer tests, and evaluation-harness integrity.