Welcome back!

Oh, hi 👋 

It’s been a minute, hasn’t it?

I’m back to curate & share insights on AI, automation, and modern Ops—for teams who don’t have a billion dollar budget to burn.

If your social feeds look anything like mine, they’re full of AI snake oil and operations fluff. I’m here to cut through that. We’ve been busy at Switchboard but I’m back to posting regularly again.

Let’s dive in.

I was reading a post about process intelligence and it made me think about trends we’re seeing with clients as well and then it hit me:

The biggest thing holding teams back from adopting AI isn’t the tech. It’s their unwillingness to rethink how they work.

🧠 AI Isn’t the X Factor. Process Is.

Why it matters:

You can’t bolt AI onto old ways of working. Traditional operating models — built for control, predictability, and static planning — weren’t designed for real-time, autonomous execution.

🔧 Workflows ≠ Work Design

The real bottleneck?

Outdated operating models — aka how your processes, roles, and decisions actually work together. If your workflows still rely on handoffs, escalations, and approvals, you’re not ready for AI.

Consider this:

• Only 12% of enterprises have embedded AI into real ops

• Most processes are too brittle, siloed, or undocumented

• Agile tech stacks exist, but teams are still stuck interpreting, not executing

🔄 What Needs to Change

Your operating model must evolve.

AI demands dynamic workflows, embedded trust, and real-time decisions. That means moving from static process maps to adaptive execution engines.

Adaptive operating models:

• Orchestrate, don’t control

• Align on outcomes, not roles

• Sense and respond, not plan and wait

• Empower humans and agents to act autonomously

🚀 The Reset Moment

You can’t optimize what you don’t understand.

To make AI work, teams need process intelligence — clear visibility into how work actually flows. That’s what unlocks agility, scalability, and measurable ROI.

💬 AI doesn’t transform your business. Rethinking your operating model does.

Read the “Break your operating model before it breaks you” report here →

AI forgets. That’s the real problem.

If you’ve ever had a long-running ChatGPT conversation or uploaded docs into a thread only to see the model “lose the plot,” you’ve experienced it firsthand: context collapse.

Why it matters: This same issue plagues AI adoption in business. LLMs can't yet think through your org's full stack of context—your workflows, docs, processes, decisions, or customer nuance—when it exceeds their metaphorical mental bandwidth.

Performance starts to get dented at 4,000 tokens on some models. By 16,000 tokens, most start to collapse. And at 60,000 tokens, they've completely lost the plot.

This plays out in real business scenarios constantly:

• Upload your 25-page quarterly financial report? The AI might remember the first half but forget critical details from the conclusion.

• Share your customer journey map and service blueprint? The model might miss connections between touchpoints.

• Need to analyze multiple department workflows together? The AI will struggle to maintain coherence across all those processes.

The result: fragmented understanding, inconsistent responses, and a frustrating experience that fails to deliver on AI's promise to streamline complex business operations.

🧪 The Big Breakthrough I’m Watching

Solving long context = unlocking AGI.

It’s not flashy, but it’s foundational. If models can’t reason across 100k+ tokens reliably, they won’t scale to your business operations. That’s why we’re tracking benchmarks like Fiction.LiveBench, built on real-world, long-story writing tasks—not search-style tests.

📊 The test asks models to:

• Track complex narrative shifts over time

• Understand character motivations and secrets

• Recall subtext spread across huge inputs

And most models fail once inputs grow too long.

🔍 New Winners & Real-World Limits

OpenAI’s o3 is finally making a big dent here.

It leads current models in comprehension at scale—crucial for dynamic, multi-layered use cases like writing, planning, and enterprise ops. It’s maintaining coherence at 120,000+ tokens.

Even state of the art models like Gemini 2.5, boasting 2.5 million token context, drop in performance by the 120k mark.

⚠️ Translation: “Supports ingesting long context” ≠ “Understands long context.”

🔄 So What?

Business AI won’t work without context mastery.

Until models can ingest, retain, and reason across everything—from meeting notes to contracts to customer interactions—AI in the enterprise will remain brittle and use cases will be strong for narrow workflows, but not company-wide usage.

💬 Long context isn’t a nice-to-have. It’s the bridge to truly intelligent systems.

What’s new:

Airtable just launched Airtable Assistant and I’m a big fan. And no, this isn’t sponsored…it’s just cool to see good product teams build (actually) useful AI features into products.

Why it matters:

This isn’t just another chat interface. Airtable has built a deeply integrated AI experience designed to enhance real business workflows. Think less fluff, more function.

Key capabilities:

• Build & query apps with natural language

• Analyze thousands of documents for structured data

• Automated web research per Airtable row

• Insights from CRM, social, support & more

How teams are using it:

• 🧾 Operations teams summarize vendor contracts to spot renewal dates or risky clauses

• 🛠️ Service-based businesses build task trackers & forms without dev help

• 📊 Sales & support teams identify top issues from hundreds of notes

• 🌐 Marketers monitor keywords and mentions using real-time web search

Under the hood:

• Easily customized and integrated with all your existing Airtable data

• Integrated web search + doc parsing

• Admins stay in control: model choice, data security, AI usage

Check it out →

Speaking of AI evolution, if you want a deep dive down where things are potentially headed, OpenAI insiders and leading experts suggest super intelligent AI is likely by 2030 — with impacts that could eclipse the Industrial Revolution.

Why it matters:

This isn’t just sci-fi.

They’ve built our a very detailed scenario called AI 2027 which maps a near-future where AI agents evolve from useful assistants into autonomous research engines that triple algorithmic progress — and eventually shape global geopolitics.

Key takeaways:

• AI starts replacing R&D: Think GitHub Copilot on steroids. Agent-class AIs redesign codebases, self-train on real-world tasks, and manage teams.

• Security gets real: As AI labs race ahead, so do fears of cyber theft, sabotage, and national power imbalances. Companies become geopolitical players.

• The tipping point: By 2027, the line between assistant and strategist blurs. Internal systems become smarter than their creators. Alignment challenges turn into national debates.

Zoom out:

For small and mid-market businesses, the message is clear: AI isn’t coming — it’s accelerating. The question isn’t if your workflow will involve agents, it’s how soon. And whether your team is set up to manage the power — and the risks.

Go deeper:

Read the full AI 2027 scenario →

That’s all for this week.

A quick question before you go:

Chat soon 👋