Thinking on AI, software, and building products.
Monday.com cut 630 jobs — 20% of staff — to become an AI work platform that does the work, not just manages it. What it reveals about AI project management.
Read the articleLoop engineering replaced prompt engineering in 2026 — Boris Cherny and Addy Osmani made it viral. But a loop that discovers, ships, and verifies needs a board.
AI-first hiring is the 2026 default: prove AI can't do the job first. But 66% of CEOs froze hiring while only 13% see org gains. The gap is coordination.
AI promised to save time. Instead, 83% of workers in 2026 say it grew their workload. Why AI work intensification is a coordination problem, and how to fix it.
An agent manager runs teams of AI agents, but most people cap out at just 4. Why span of control — not smarter models — is the real limit on agents in 2026.
Gartner says agentic arbitrage puts $234B of SaaS spend at risk by 2030 as AI agents work through APIs, not seats. Why per-seat project boards die first.
Anthropic ships 8x more code, but coordination didn't scale. In 2026 the AI bottleneck moved from writing code to coordinating it. Why more PMs won't fix that.
OpenAI's June 2026 Codex study found non-developer AI use up 137x as work shifts from consultation to delegation. When everyone delegates, who tracks it?
Section's July 2026 report found fewer than 10% of workers can even define an AI agent. The AI proficiency gap is an interface problem, not a training one.
OpenAI's ChatGPT Work made the AI super app real — one agent that ships finished work in hours. But a team of agents and humans still needs a shared board.
RTO mandates cite collaboration — 68% of workers agree. But AI now automates that collaboration, and your AI agents can't come in. Coordination isn't a place.
Anthropic's July 2026 data shows 91% of Claude Cowork isn't coding — it's the office 'work around the work.' Why AI agents doing it still need a shared board.
The AI layoff boomerang is here: firms cut workers citing AI, and 55% now regret it. Why replacement keeps backfiring — and why rehiring isn't the real fix.
Shadow agents — autonomous AI running under employee logins that IT can't see — are 2026's fastest-growing work risk. You can't audit your way out of it.
In July 2026, Microsoft and AWS put $3.5B into forward-deployed engineers to make AI actually ship. The real bottleneck isn't the models — it's coordination.
Zuckerberg says Meta's $145B AI agent bet stalled after four months. Here's why AI agents fail at scale — and why the fix is coordination, not a bigger model.
BCG research says labeling AI agents 'employees' makes managers catch fewer errors and offload the blame. The 2026 org-chart trap — and how to avoid it.
Meta ran keystroke surveillance on 6,500 engineers and got a revolt. Why activity tracking can't see real work — and what does once AI agents join the team.
Botsitting is the 6.4 hours a week workers now spend babysitting AI. A June 2026 study shows why it eats your time savings — and where the real fix lives.
Google found more AI agents make sequential work 39-70% worse, not better. Microsoft says org factors drive 67% of AI impact. The variable is coordination.
New 2026 benchmarks show AI agent teams score up to 50% lower than a single agent. The failure is coordination, not coding — and a shared board fixes it.
Microsoft says every worker is an 'agent boss' now. A June 2026 survey found 72% run AI agents with unmanaged risk. The fix isn't chat — it's a shared board.
PwC's 2026 AI Jobs Barometer (June 15) found the most AI-exposed firms grew headcount 52% — not less. The dividing line isn't AI access. It's coordination.
Microsoft's 2026 Work Trend Index named the Transformation Paradox: workers are AI-ready, their systems aren't. Agent use grew 15x; only 19% capture the value.
The AI productivity paradox: agents now do 26 minutes of solo work per session, yet only 13% of workers see company-wide gains. The fix is coordination.
PwC's 2026 AI Jobs Barometer found a new divide: companies that amplify human expertise with AI outpace those that automate. It's a board, not a slide.
Microsoft is pulling Claude Code by June 30. Uber's CTO admits the 2026 AI budget is gone. Token bill shock is here — and routing it is a board problem.
AI-washing layoffs: 59% of hiring managers admit they emphasize AI because it plays better. A 2025 CMU study finds agents fabricate to mask quality gaps.
Claude Fable 5 launched June 9, 2026 at twice Opus pricing, with built-in risk routing. Frontier AI is now a rationed resource — routing it is a board problem.
Atlassian's State of Teams 2026 named the AI fragmentation tax: $161B a year on the Fortune 500, 6.4 hours per person per week lost to coordination chaos.
Block cut 4,000 jobs. Amazon cut 14,000. Gartner forecasts half of middle management gone by year-end. Gallup's 2026 data shows where the coordination work went.
ClickUp cut 290 humans, deployed 3,000 AI agents, and called it the 100x org. Gartner says replacement doesn't pay off — amplification does. Here's the data.
AI coding agents now top 80% on SWE-Bench Verified — but merge in production at roughly half the rate of human pull requests. The 2026 benchmark-to-merge gap isn't a capability problem. It's how 'done' is defined.
AI agent identity went GA in 2026 — Microsoft Entra, Google's Agent Registry, NIST initiative. Identity tells you who. The board tells you what shipped.
AI agents now run autonomously for 30+ hours, with task horizons doubling every four months. The two-week sprint can't see the work anymore. Here's the fix.
Agent observability hit 89% adoption, but evals lag at 52% — and Stanford finds most pilots never ship. The fix is a board where 'done' is a verifiable assertion.
The 2025 DORA report found AI amplifies whatever's already in your engineering system — for better or worse. Your project board is where the truth shows.
Spec-driven development is the 2026 successor to vibe coding: a versioned spec is the source of truth and AI coding agents generate the rest from it.
A 2025 Harvard Business Review study named the AI-generated work that looks finished but isn't: workslop. Forty percent of workers said they'd received it in the past month, and each instance cost nearly two hours to clean up. The fix isn't less AI — it's a shared board where 'done' has to be earned.
Gartner expects the average Fortune 500 to run over 150,000 AI agents by 2028 — up from fewer than 15 in 2025 — while almost no one can produce an inventory of the ones already running. Agent sprawl is duplicated work, conflicting outputs, and zero visibility. The fix isn't another governance dashboard. It's a shared board agents work on directly.
Gartner is warning buyers about 'agent washing' — old automation rebranded with the word agent stitched onto the label. Project management is where it hides best. Here's how to tell a real AI teammate from a dressed-up rules engine, and why the difference shows up on the board.
Background AI coding agents run while you context-switch and deliver a finished pull request when they're done. The productivity is real — but the moment work moves off your screen, you lose visibility. The fix isn't more chat. It's a shared board agents operate on directly.
Gartner predicts over 40% of agentic AI projects will be canceled by 2027. The teams running agents in production know why: multi-agent systems don't fail inside any single agent — they fail at the handoff. The fix isn't a smarter agent. It's a shared board.
AI contract review crossed into the mainstream this year — most in-house legal teams now use it or are evaluating it. But clause-checking is the easy part. The real shift is the agreement layer: contracts that draft, sign, and keep pace with the agents now doing the work.
The average manager spends five hours per week on status communication. AI-generated reports deliver better stakeholder visibility in thirty seconds — with a shareable link, not a meeting invite.
AI project management tools are only as good as the data on the board. Five default fields capture five percent of the picture. Custom metadata is the missing layer that makes AI actually useful.
Over 90% of businesses use AI. Most cannot point to measurable productivity gains. The gap between adoption and impact is not a technology problem — it is an architecture problem.
AI coding tools made individual developers faster. They did not make teams faster. The bottleneck moved from writing code to coordinating work — and that is a project board problem.
Solo founders are shipping products that used to require teams. The secret is not working harder — it is delegating to AI agents that handle code, design, marketing, and ops. Here is the playbook.
Sprint planning meetings burn twenty engineer-hours to produce a backlog that looks like whatever was at the top of Jira. AI is about to change that — not by replacing the meeting, but by eliminating the need for it.
Every week someone uses ChatGPT to generate a sprint plan. Three days later, none of it is tracked. The gap between planning and managing is where teams lose weeks — and where purpose-built AI PM tools change the game.
The Model Context Protocol gives AI agents a universal way to connect to every tool. But connections without coordination is chaos. Here is why the missing piece is not another API — it is a new kind of project management.
Most AI agents start from zero every session. In project management, that means re-explaining your team, your priorities, and your decisions every single time. The agents that break through are the ones that remember.
A new class of company is emerging — not ones that adopted AI, but ones where agents are embedded in every operational layer from day one. Their org charts, toolchains, and economics look nothing like their predecessors.
Copilots made individual developers faster. Agents make entire teams faster. But only if your coordination layer can keep up. Here is what the shift from assisted coding to autonomous execution actually looks like — and why project management is the missing piece.
The average enterprise runs 12 AI agents. Half work alone — no shared context, no coordination, no audit trail. Capability is not the bottleneck. Governance is.
Your codebase is growing faster than your team can understand it. AI writes clean, passing code — but nobody absorbs it. The gap between code volume and human understanding is the most dangerous debt you are accumulating.
Traditional wikis decay the moment they are written. When AI agents are part of your team, stale documentation does not just waste time — it causes wrong decisions at machine speed. The future is documentation that generates itself from the work.
95% of developers use AI coding tools weekly. Almost none trust AI to manage their projects. The gap is not about capability — it is about verification. Here is how to close it.
The AI agent market is growing 46% annually. But the real story is small teams using agents to ship at a pace that makes competitors assume they have ten times the headcount. Here is the playbook.
Enterprise interest in multi-agent systems surged over 1,400% this year. Single agents hit a ceiling — real work needs specialized agents coordinating on shared state. The orchestration layer that makes this work already has a name: project management.
The teams getting real results from AI aren't the ones with the best models. They're the ones that figured out what to feed the model before the conversation starts. Context engineering is quietly becoming the discipline that separates magical AI products from generic ones.
Eighty percent of developers now use AI coding agents daily. But trust in AI output is falling even as adoption rises. The bottleneck isn't the agents — it's the coordination layer that doesn't exist yet.
Traditional automations save clicks. AI-powered automations make decisions. The line between automation and agent is dissolving, and the teams that get this right will ship twice as fast.
40 percent of enterprise apps will include AI agents by year-end. Most teams have no idea how to manage a workforce that is half human and half AI. Here is what actually works.
Every PM tool shipped an AI feature. Most bolted it onto an existing interface. A few rebuilt around it. Here is where AI project management actually stands — what works, what is marketing, and what matters.
The AI project management market is headed to $13 billion. But most tools still assume a human is staring at a dashboard. The real shift is agents that manage projects — not just assist with them.
When anyone can generate working code with a prompt, the bottleneck isn't writing software anymore. It's knowing what to build — and that's a project management problem.
Building got an order of magnitude cheaper. Buying didn't. The old framework for when to build and when to buy needs an update.
Autonomous agents are joining project boards alongside humans. Your PM tool needs an API that treats them as first-class participants, not an afterthought.
No standups, no sprint planning, no retros. We replaced every recurring meeting with systems that make information flow without human intermediaries.
The difference between a product you tolerate and a product you love is not features. It is a thousand small decisions made by someone who cared.
The dirty secret of PM software: most teams abandon it within three months. The problem isn't the features — it's the model.
Everyone says agents are cheap. They are — per task. But 'cheap per task' and 'cheap to run' are different claims. Here are our real numbers.
Most companies experiment with agents in engineering. The interesting question is what happens when every department uses them.
A settings page is the product admitting it doesn't know what you need. We replaced ours with a conversation.
When AI handles 80% of the mechanical coding, charging by the hour means billing for time you didn't spend. Fixed-price is the future.
Onboarding wizards are a confession that your product doesn't know what users need. We replaced ours with a single question: what are you working on?
The average PM tool has 47 configuration options before you create your first task. We think that number should be zero.
What happens when a client's journey from 'I'm interested' to 'the project is done' runs through one system with zero handoffs.
Agents don't create a productivity problem. They create a coordination problem. Here's how the best teams solve it.
Most APIs are built for frontends. Agents need explicit state machines, atomic operations, and structured errors.
PRs were the best review tool for human-only teams. Agents need something better.
Every PM tool starts by giving you a board. We think that's the problem.
How a single repository becomes the foundation for an entire company's software.
What happens when the chat is the project, not a sidebar to it.