2025-09-15 08:00:00
Now that we’ve compressed nearly all human knowledge into large language models, the next frontier is tool calling. Chaining together different AI tools enables automation. The shift from thinking to doing represents the real breakthrough in AI utility.
I’ve built more than 100 tools for myself, & they work most of the time, but not all the time. I’m not alone. Anthropic’s Economic Index report reveals that 77% of business use of Claude centers on full-task automation, not co-piloting.
Anthropic published documentation last week about token efficiency & re-architecting tools to optimize their use. The guidance was counterintuitive : instead of many simple tools with clear labels, create fewer, more complex tools.
Here are the seven email tools I built - Ruby scripts, each with a clear purpose. The “Safe Send Email” script was designed to prevent the AI from sending emails without approval.
draft_email.rb
send_email.rb
forward_email.rb
find_and_draft_reply.rb
read_email.rb
archive_emails.rb
safe_send_email.rb
Beautifully naive, simple, & clear, Shouldn’t a language model be able to read these & know exactly what I was asking it to do? But it’s not this simple!
Anthropic recommends creating complex tools. Their research shows that “requests save an average of 14% in output tokens, up to 70%" when using sophisticated, parameter-rich tools instead of simple ones. The reason? AI systems understand full context better than fragmented intent.
I spent the weekend consolidating all my tools into unified tools, like this one for email:
ruby unified_email_tool.rb \
--action send \
--to "[email protected]" \
--subject "Q4 Strategy Review" \
--body "..." \
--cc "[email protected]" \
--format concise
The impact on accuracy was immediate. Claude’s success rate approaches 100%. The system is faster. As a result, I’m using far fewer tokens with a more efficient system.
Here’s my current mental model:
| People Need | AI Systems Need |
|---|---|
| Cognitive chunking | Complete context |
| Progressive disclosure | Parameter-rich interfaces |
When I redesigned for AI cognition rather than human intuition, everything improved. My CRM operations, calendar management, & database workflows all became more reliable when consolidated into comprehensive, parameter-heavy tools. Accuracy improved, so the total cost was reduced significantly.
But don’t ask me to use the tools. I’m now a bit lost amidst the complexity. This is an inevitable corollary of working at higher levels of abstraction, no longer deeply understanding the machine.
We spent decades making software simple for people. Now we’re learning to make it complex for AI.
2025-09-12 08:00:00
Like breakfast at a diner, software is cheap & fast to make. Ask for a new task management tool & you’ll have the first version in less time & for less money than an omelette.
AI-built tools may not last long. Some survive only a few minutes, long enough to answer “What’s our turnaround time this week?”
Others remain useful for a few days or weeks, like an app that can spin up a lightweight project tracker for onboarding Walmart. Sometimes they persist beyond a month.
If permanence defined the last two decades of software, impermanence may define the next.
We now see three layers forming along a continuum:
Durable SaaS: long-lived systems of record like a pipeline dashboard.
Ephemeral Apps: short-lived tools like the Walmart onboarding project tracker.
Instant Questions: one-off queries such as “Tell me about the Apple account.”
By orders of magnitude, ephemeral apps & instant questions will outnumber SaaS applications, perhaps millions to one. The dopamine high of instant problem-solving is addictive & accelerates careers.
Underpinning all of these applications will be a system of record, often an existing platform but increasingly a new one.
Finance & operations teams depend on persistent dashboards for governance & reporting, as if the definition of a metric could be standardized across the organization. Marketers spin up data apps that last a few months to analyze their paid spend performance, & support teams ask quick questions about turnaround time before moving on.
The trusted models, permissions, & business logic of an underlying BI platform like Omni give users confidence to experiment with ephemeral apps & instant questions. Control of the foundation ensures control of the layers above.
This design pattern extends beyond AI to most other software.
Whether you’ll have a coffee, an omelette, or a five-course meal, the next great startup will serve it all.
2025-09-11 08:00:00
Understanding technology is at the core of how we research new theories. As part of that effort, Theory is starting a developer relations team.
Developer relations for venture capital? What does that mean?
Internally, we’ve built hundreds of agents, we’ve struggled to debug tool calls, tested large action models, & work through AI every day. We’ve hosted events on how to build with AI.
We’d like to contribute back to the community with our experiences & learn from other builders.
All of our workflows of the last decade have suddenly changed. Our broader community is exploring what new workflows are possible & which are best. This new effort is another step in that direction.
So if you see Mischa, our new developer relations lead, at an event, say hi! He’ll have plenty to share!
2025-09-09 08:00:00
MCP enables an AI to understand which tools rest at its disposal : web search, file editing, & email drafting for example.
Ten months later, we analyzed 200 MCP tools to understand which categories developers actually use.
Three usage patterns have emerged from the data :
Development infrastructure tools dominate with 54% of all sessions despite being just half the available servers. Terminal access, code generation, & infrastructure access are the most popular.
While coding, engineers benefit from the ability to push to GitHub, run code in a terminal, & spin up databases. These tools streamline workflows & reduce context switching.
Information retrieval captures 28% of sessions with fewer tools, showing high efficiency. Web search, knowledge bases, & document retrieval are key players. These systems are likely used more in production, on behalf on users, than during development.
Everything else including entertainment, personal management, content creation, splits the remaining 18%. Movie recommenders, task managers, & Formula 1 schedules fill specific niches.
MCP adoption is still early. Not all AIs support MCP. Of those that do, Claude, Claude Code, Cursor top the list (alliteration in AI). Developer focused products & early technical adopters are the majority of users.
But as consumer use of AI tools grows & MCP support broadens, we should expect to see a much greater diversity of tool use.
2025-09-08 08:00:00
$4b in revenue run-rate. There are two data giants both now at this mark after Databricks announced it surpassed the threshold.
This is an opportunity to compare the two leading data companies at a revenue intersection.
| Metric | Databricks | Snowflake |
|---|---|---|
| Revenue Run Rate | $4.0B | $4.1B |
| $1M+ Customers | 650 | 654 |
| Net Dollar Retention | 140% | 125% |
| YoY Growth Rate | 50% | 28% |
| Valuation | $100B | $75.9B |
| Market Status | Private | Public |
| AI Revenue | $1B | Not disclosed |
Both are at $4b in revenue. Each claims over 650 customers paying $1M+ annually. Each boasts strong net dollar retention (140% vs 125%).
Databricks is growing at 50% compared to Snowflake’s 28% & in the private market trades at a premium for that growth rate. Snowflake has reaccelerated but it was about a year later than Databricks.
Comparing the valuation to revenue-run-rate shows Databricks is trading at a 35% premium to Snowflake.1
Databricks is valued at $100B privately while Snowflake trades at $75.9b publicly. In this market, every 1% of growth adds 0.3x to valuation multiple. Given Databricks’ 22-point growth advantage, the 35% premium may actually understate the true difference in ultimate business size.
This premium reflects the scarcity of high-growth data platforms in public markets. There really is no equivalent to Databricks today. Palantir at 39% growth trades at 75x forward (not run-rate). Rubrik, in the twilight of a transition from on-prem to cloud, trades at 15x forward on 44% growth.
The 35% valuation premium reflects both Databricks’ superior growth & the market’s bet on AI. With AI revenue already at $1B & driving concomitant compute demand, Databricks has positioned itself at the center of the most valuable trend in enterprise software.
$100b/$4b = 25x vs $75.9b/$4.1b 18.5x ↩︎
2025-09-04 08:00:00
Each morning brings the excitement of a world that seems to have changed overnight.
A new model arrives promising a novel way of working. Like unboxing an iPhone in 2008, but with the steady cadence of the near-daily Amazon Prime delivery ferrying its promise of expected newness.
Those gifts present puzzles for those on the cutting edge. They offer the opportunity to look at a problem with a supercharged machine, gleaming tools replacing the loved, blunted ones.
The puzzles contain at their center the promise of a fortune : rapidly appreciating stock & time saved, the promise of a better future.
AI is reimagining how we work. In San Francisco bars, the ultimate flex has become regaling acquaintances with tales of robotic workers acting on every whim, a testament to the power at our fingertips.
At the next cocktail table, a different crowd shares the wistfulness of a career that has suddenly been upended.
For many, AI represents a Pandora’s box of unwelcome change. The determinism of steady careers & processes has been replaced by the unpredictable agents that have brought the efficiencies of Taylor’s interchangeable parts to knowledge work.
AI reinforces imposter syndrome. We look around to find those who know what we’re doing with AI. How should I prompt this impossibly complex machine cloaked in a simple text box? How do I guide a team through a fog of rapid change?
At the dinner table, looking at bright young faces drinking their fortified milk wiping their mustaches, we wonder what world they’ll inherit from this transformation. What education could possibly prepare them?
Steven Jay Gould would dub it the end of the punctuated equilibrium. AI has upended the pax technologica of the last ten years, forcing rapid evolution for all of its opportunity both new & lost.
