Human Touch Required for True Accessibility

Most accessibility issues are not found by automated scans. And the ones that matter most are almost never caught. Teams run a scan, see a score, fix a handful of flagged issues, and assume they are in a reasonable place. From a reporting standpoint, that looks like progress. From a risk standpoint, it is often meaningless. Automated tools are good at finding surface level issues. Missing alt text. Color contrast problems. Basic structural gaps. They do not tell you whether a user can actually complete a task. Can someone navigate your primary workflow using only a keyboard. Does focus move in a predictable way. Do dynamic elements announce themselves correctly. Can a form be completed without hitting a dead end. That is where real accessibility lives. And that is exactly where most tools stop. I have seen platforms with strong automated scores completely fail when used with a screen reader. Not edge cases. Core functionality. This creates a dangerous gap. Teams believe they have coverage because they have data. In reality, they have partial visibility at best. Accessibility is not a checklist. It is an interaction problem. If your process ends with a scan, you are not validating usability. You are validating that a tool ran successfully. Those are not the same thing. The question is not whether issues exist. It is whether they show up before or after your users find them. That distinction matters more than the score.

Modern web experiences fail when accessibility isn’t treated as a core condition of how systems are built. As AI accelerates development and digital platforms scale faster than governance can keep up, accessibility gaps become structural problems, not bugs to fix. The teams building resilient, future-ready technology are the ones embedding accessibility directly into their engineering practices, design systems, and decision-making—because end user access, and organizational reach, depends on it. Our article for WIRED explains how accessibility functions as essential infrastructure in modern digital ecosystems, and why tech leaders can’t afford to treat it as anything less. https://ow.ly/m2pT50YH8Ij #TechLeadership #InclusiveDesign #DigitalAccessibility #FutureOfTech #UserExperience #AccessibleTech

Claude just made things more interesting. Now you can send tasks to Claude from your phone and let it work through your paired desktop. That means you can start something while sitting anywhere and keep track of it from one conversation across both devices. This feels like a big step for people who want to manage real work from mobile without being stuck at a desk all the time. If this works smoothly in real use, it could make phone-to-desktop workflow much easier for research, writing, coding, and task handling. The real question is not just convenience. It is trust, control, and how safely desktop access is handled. Useful update. But security and permissions will matter a lot. #Claude #AI #ProductUpdate #Automation #Tech #Productivity

Follow-up to last week's piece on Copilot Cowork custom skills. Now that the subfolder pattern works, the next question is how to actually design skills that don't rot after the third one. Short answer: SKILL.md isn't the skill. It's a dispatch router. Your format templates, quality checklists, and worked examples belong in subfolders where the right team can own them. Full post uses a Change Advisory Board submission skill as the worked example. Also covers portability across the wider Agent Skills ecosystem (same format now runs in Claude, Copilot, ChatGPT, and Codex) and a warning about public skill marketplaces. #Microsoft365 #Copilot #M365Architecture #EnterpriseAI

I built a Claude-powered desktop execution system for my Conigma team. I'm giving it away for free. Most people use Claude like a fancy chat window. The smarter play is using it to handle the delivery work so your team focuses on outputs that actually move things forward. I built a system with: • 7 Cowork capabilities • 1 master setup framework that makes all of them compound This isn't surface-level stuff. It's a full local execution system covering: • Local file access (no limits, no copy-pasting, work lands in your folder) • Persistent memory (saves to your machine, no cap, gets smarter every session) • Tools and connectors (Gmail, Google Drive, Google Calendar, Notion in one workflow) • Skills (saved repeatable workflows that run the same process every time) • Cowork Projects (autonomous task execution with memory that improves week over week) • Scheduled tasks (workflows that run before you sit down) • Browser extension (form filling, page navigation, click-through tasks) Each capability handles one job. And it handles the repetitive parts better than burning hours manually copying outputs, re-uploading files, and re-briefing Claude from scratch every session. Instead of running one-off prompts and doing the work yourself... You get a Claude-powered local execution system you can plug straight into your existing workflows. I packaged the whole thing. All 7 capabilities. The setup guide. The settings. The skill-building process. The scheduled task framework. Want it? Connect + Comment "COWORK" I'll send it over.

Deploying Virtual Agents into production takes more than a good demo. When agents handle sensitive, business-critical workflows such as payments, health information, and insurance claims, the challenge is not just building them, it is making sure they operate safely, predictably, and at scale. At Level, we are focused on giving Agent Designers and Business users robust tools to design, control, and evaluate Virtual Agents in real production environments. A production-grade Virtual Agent should be able to: - Resist malicious attempts to hijack instructions or override policies - Prevent leakage of sensitive information, such as private data or internal API names - Protect against unauthorized tool access without proper authentication - Enable teams to understand and evaluate agent quality at scale after rollout Solving this requires a multi-layered approach: - Agent harnesses that support deterministic execution blocks for tighter control over critical workflows - Progressive tool disclosure so sensitive tools are accessible only after proper authorization - Default input and output guardrails to reduce malicious or unsafe behavior - Robust post-conversation quality assurance to help Business users detect, diagnose, and improve failures at scale. Join me and Rob Dwyer for a webinar on May 7 @ 1:00 pm ET. We will discuss what it takes to move from promising agent demos to production-ready Virtual Agents — with practical approaches for safety, control, authorization, and quality evaluation. https://lnkd.in/gWaPi79U

𝗧𝗵𝗲 𝗜𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝗰𝗲 𝗢𝗳 𝗔𝗰𝗰𝗲𝘀𝘀𝗶𝗯𝗹𝗲 𝗧𝗼𝗮𝘀𝘁 𝗡𝗼𝘁𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 You want to make your toast notifications accessible to all users. This means considering screen reader users, keyboard users, and assistive technologies. Here are some key accessibility goals: - Notifications should be announced to screen readers - Users should have enough time to read the message - Keyboard users should get the same behavior as mouse users - Users should be able to manually dismiss notifications To achieve these goals, you can use live regions. This tells screen readers to announce updates when the user is idle. You can also pause the timer when the user interacts with the notification. Each toast should include a dismiss button with a clear label. This ensures screen readers clearly communicate the purpose of the button. Notifications should appear visually without stealing focus. This prevents interruptions while users are typing or navigating through the interface. Supporting multiple notifications is also important. This allows multiple notifications to appear without overlapping content. Accessibility matters because poorly implemented toast notifications can create problems for assistive technology users. By incorporating ARIA live regions, you can ensure notifications remain usable and understandable for all users. Source: https://lnkd.in/gDe3mUHR

Why Automation isn't just about Speed—it’s about Strategy. In the early stages of building a business, "doing it all" is often worn as a badge of honor. But as we scale, the manual processes that once kept us afloat can quickly become the anchors that hold us back. At Improvised Tech, we believe that the true power of AI and Full-Stack automation isn't just in doing things faster; it’s about freeing up the human mind to focus on high-level decision-making and creative problem-solving. We aren't just building software; we are building ecosystems. Whether it’s integrating seamless payment gateways or deploying AI-driven business operating systems, our goal is to turn complex technical debt into streamlined operational assets. The Improvised Tech approach focuses on: Precision Engineering: Moving from generic "web design" to robust Website Development that serves as a functional tool, not just a digital flyer. Systematic Growth: Implementing AI-driven solutions that scale with your ambitions, not against your resources. Global Standards, Local Context: Solving modern business challenges with a deep understanding of the professional landscape. Innovation isn't an event—it’s a commitment to continuous improvement. Let’s build something that lasts. #ImprovisedTech #TechInnovation #FullStackDevelopment #AIAutomation #BusinessSystems #SoftwareEngineering #FutureOfWork

As systems shift from executing commands to interpreting intent and orchestrating work, the way we think about control is starting to break down. The common reflex is familiar: keep a human in the loop. It sounds safe, responsible—even obvious. If a person reviews the output, we assume the system remains under control. But that assumption hides a deeper design problem. Human-in-the-loop (HITL) is often treated as a solution when it is, in practice, a placeholder. It inserts a human into the process without defining who is responsible, when intervention is required, or how authority holds under scale, cognitive load, and #distributedwork. As a result, control doesn’t fail all at once—it fragments. #Responsibility diffuses. #Decisions become harder to trace, not easier to govern. This is not just a #usability issue. It is a product strategy issue. Because the moment systems act on our behalf, we are no longer designing interfaces—we are designing how responsibility, authority, and oversight are structured across people and machines. So the question is no longer: “Is a human in the loop?” But: * Who is actually accountable for this decision—and how is that made explicit in the system? * Where does oversight break down under scale, cognitive load, or coordination—and how have we designed for that? * Which role is expected to act at this moment—and do they have the context and authority to do so? Human-in-the-loop promises control. But without clear roles and boundaries, it creates the illusion of it. If this tension feels familiar, I explore it in more detail here. As I wrote this post, I thank the thought leaders that came before me: Ben Shneiderman, Saleema Amershi, Daniel Weld, Mihaela Vorvoreanu, Adam Fourney, Besmira Nushi, Penny Collisson, Alexander D'Amour, Katherine Heller, Dan Moldovan, Abe Crystal, Elizabeth Grigg-Ellington, Daniel Diaper, Neville Stanton, John Lee, Katrina See, Raja Parasuraman, Chris WICKENS, Dr. Michael Vössing, Niklas Kühl, Matteo Lind, Gerhard Satzger, Zana Buçinca, Maja Malaya, and Krzysztof Gajos. If you want to learn more about Human-Agent Centered Design, check out the website: https://hacd.lovable.app/ The evolution of this project is ongoing, with continued collaboration from André Neves, Silvio Meira, Filipe Calegario, Marcello Bressan, Diego Credidio, Farley Fernandes, César França, Fernando Mattoso Lemos, Rui Belfort, and Rodrigo Medeiros. #ArtificialIntelligence #AgenticAI #UserExperience #DesignProcesses #UXforAI #AI #AgenticAI #HumanAgentCenteredDesign #ProductStrategy #HumanInTheLoop #HITL #UXDesign #DesignLeadership #CSCW #ComputerSupportedCooperativeWork

Is your accessibility program caught between the slow "manual trap" and the inaccurate "automation myth"? Relying solely on manual audits is too slow to scale, yet overreliance on "AI-only" tools and overlays can miss over 50% of critical issues, leaving you exposed to legal and compliance risk. We need a modern approach to web accessibility. In our latest blog post, Foulk Consulting Performance Expert, Caleb Billingsley, breaks down why the most effective programs don't choose between artificial intelligence and human expertise – they combine them. Caleb outlines the "𝗕𝗲𝘀𝘁 𝗼𝗳 𝗕𝗼𝘁𝗵 𝗪𝗼𝗿𝗹𝗱𝘀" model: ⚡ Using AI for speed, scale, and prioritization. 🧠 Grounding that work in 150 combined years of human judgment, manual validation, and assistive technology testing for accuracy and real-world usability. Stop settling for static audit reports or unreliable overlays. Learn how to move faster toward WCAG conformance and reduce risk without sacrificing quality engineering discipline. 𝗚𝗲𝘁 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝗶𝗻𝘀𝗶𝗴𝗵𝘁𝘀 𝗵𝗲𝗿𝗲: https://lnkd.in/eFcNpDg3 #FoulkConsulting #WebAccessibility #WCAG #DigitalQuality #QualityEngineering #AIinTesting #InclusiveDesign #SoftwareLifecycle