If your team is still wedging AI into workflows you designed for humans, you're solving the wrong problem.
This week's edition covers:
- Kevin's Take: what Notion's "spec-driven development" tells you about how to rewire your marketing team for AI
- The Signal: why OpenAI's new "Deployment Company" is a bigger story than any model release, and what it confirms about top-down AI adoption
- Tools & Tactics: an audience research playbook for mid-market RevOps, plus a public library of 1,000+ Claude/Codex agent skills your team can grab today
- Quick Links: the strongest argument that AI won't take your job, the 2026 GTM survey worth filling out, Moore's Law's broken economics, and whether agentic commerce is real
Kevin's Take
Notion's engineers don't really write code anymore
I was listening to Claire Vo interview Ryan Nystrom from Notion's engineering team this week, and something he said stuck with me. The way Notion does engineering now is what Ryan calls spec-driven development. Engineers voice-dictate the idea using Whisper, AI formats the dictation into a proper spec document, the spec gets committed to the repo the same way code used to, and then agents go off and implement the feature. Pull requests come back inside of 20 minutes with screenshots attached. Their internal system called Boxy lets an engineer mention Codex inside a GitHub comment and get a full PR back the same morning. The line Ryan used that captures the whole thing is, "agents do the coding while you do the thinking."
That's a very different operating model than what most marketing teams I see are running. Most marketing leaders right now are putting Claude or ChatGPT into the same workflow they were running a year ago, using it to write the draft faster, to summarize the meeting faster, to polish the deck faster. That's fine, and it produces real efficiency, but the actual shape of the work hasn't changed at all. You've just compressed parts of it.
What Notion is doing is something else, and I think this is the pivot all of us are going to have to make at some point. They've rebuilt the workflow so the human writes the spec and the agent does the execution. The bottleneck used to be execution capacity, you can only ship so many features per week with so many engineers on staff. Now the bottleneck is the quality of the spec. If the spec is vague or under-specified, the agent produces something confidently wrong, because a model doesn't stop and ask you what you meant the way a junior engineer would. It just builds whatever the words said.
At Mighty and True we've been working on this exact pivot, although on the marketing side rather than the engineering side. The newsletter you're reading right now is produced by a pipeline I built using Claude Code, Supabase, Vercel, and Ghost. The discovery and scoring of articles happens automatically every morning, the take is drafted by Claude Opus against a voice rules document, and the whole edition gets stitched together and pushed into Ghost on a Friday cron. My job used to be "write the newsletter every week." Now my job is to write the spec the pipeline operates on, the voice rules, the curation feeds list, the scorer rubric. When the pipeline drifts off voice, it's because the spec needs sharpening, not because the agent failed. (That's actually what happened with this very edition, by the way, the spec broke down somewhere upstream so I'm hand-editing this week, which is itself a good lesson in why the spec matters.)
If you're running a marketing team right now, my recommendation is to pick one repeating workflow that's eating your team's time, your campaign brief process, your ABM target list, your monthly competitive intel scrub, and try to rewrite it the Notion way. Instead of asking where Claude could slot into the current process, you ask what spec a capable agent would need to do this work without you in the loop. Most teams will discover the spec doesn't actually exist anywhere, your team has been operating on tribal knowledge and unwritten taste calls, which is exactly why this work has been impossible to delegate to an agent so far. Writing that spec is the work.
It's a real game changer once you make the pivot. Talk soon.
— Kevin Kerner, CEO, Mighty & True
The Signal
OpenAI is forming a "Deployment Company" and it's a bigger story than any model release | Ben Thompson, Stratechery (3 min read)
Ben Thompson argues this week that OpenAI is spinning up a new entity whose job is not to build models but to deploy them inside large enterprises, and the other AI labs are not far behind. His thesis is that the impact of AI inside companies is going to require top-down implementation, because the kind of operating-model rewiring described above doesn't happen by accident. The companies that win the next 24 months are going to be the ones whose leadership has the executive will to rebuild workflows around AI, rather than the ones racing to deploy the latest model.
Why it matters: This is the same insight playing out at Notion at the engineering layer and at Mighty and True at the marketing layer. If your AI strategy is currently described as "let each team pick the tools that work for them," you've handed the most important operating-model decision your company will make this decade to whoever happens to be the most curious person on staff that week. The leaders getting this right are picking one workflow at a time and rebuilding it deliberately.
Tools & Tactics
🎯 How we'd market to mid-market RevOps leaders | Amanda Natividad, SparkToro (3 min read)
Amanda walks through the actual audience research SparkToro would do for a hypothetical mid-market RevOps campaign, which sources they would pull from, which questions they would ask, which signals they would ignore. It's a methodology piece rather than a playbook, and it's useful because it shows the thinking most "audience research" decks gloss over.
Why it matters: If your team is writing campaign briefs without a documented audience research step in front of them, you'll run into the same spec problem that breaks marketing AI workflows. Agents (or humans) cannot act on briefs that assume audience clarity that hasn't actually been built. Use Amanda's framework as a template for what good looks like before you write your next campaign brief.
🛠 awesome-agent-skills: a public library of 1,000+ agent skills | VoltAgent, GitHub
A community-maintained collection of agent skills built by dev teams and shared on GitHub. The skills work with Claude Code, Codex, Gemini CLI, and Cursor. Most are engineering-flavored, but the marketing-adjacent ones (SEO audits, competitive intel) are usable today.
Why it matters: If you're trying to stand up your own agent stack for marketing workflows, this is a much faster starting point than writing skills from scratch. Fork the closest one to what you need, modify it for your context, then deploy it inside your tooling. The same logic as not building a CRM from scratch, most of what your team needs has already been built by somebody else.
Quick Links
Worth Reading
- The best argument I've heard for why AI won't take your job: Casey Newton interviews Box CEO Aaron Levie. Levie's argument is that jobs will survive but they will look very different from how they look today. (via Casey Newton)
- 2026 Theory GTM Survey: Tomasz Tunguz is running his annual 25-question survey for B2B startups, and respondents get the aggregated benchmarks back when it closes. About five minutes to fill out. (via Tomasz Tunguz)
- The broken bargain of Moore's Law: Azeem Azhar argues the physics-driven slowdown is now being matched by economics catching up, which means compute will not keep getting cheaper at the rate most marketing budgets have been assuming. (via Azeem Azhar)
- Is agentic commerce an oasis or a mirage?: AdExchanger weighs how ready consumer-side AI agents actually are to buy on people's behalf today. The honest answer is mostly mirage for now, but the underlying surfaces are getting built fast. (via AdExchanger)
Strategic Gap Blueprint: a 45-minute session to find the gaps in your marketing
Book a 45-minute working session with the Mighty and True team. We dig into your current marketing strategy, surface the gaps in your approach, and come back with a written analysis on what to fix and where to invest. There's no pitch attached. Just an honest read on what's working and what isn't.
▶ Book a Gap Blueprint session
Join the group we're building
We're building something for marketing leaders who are tired of vendor-sponsored best practices content. A small group. Real conversations. No pitching. If that sounds like your kind of thing, hit reply and I'll share the details.