10 New Software Capabilities
Announcing a change to the theme of this newsletter and reviewing the 10 new software capabilities that AI has unlocked.
It's been an eventful two and a half years since my last post for this newsletter:
We sold our startup Laskie, a job matching platform, to Twitter/X and I spent 18 months reporting to Elon. Yes, I have stories.
I worked on several interesting projects at X where I got to build and deploy AI features at a massive scale.
I started TimeToBuild, an AI side-project accelerator, where I helped 20 teams go from zero to one on AI-focused products. Several of those teams are doing quite well.
I’ve logged a few thousand hours building AI-powered products using AI-powered tools.
I've gone deep on AI and I've never been more pumped about a technology in my 20-year career. It will unlock better products, better experiences, and unprecedented business automation.
That’s why I’m rebranding this newsletter from all the other things to Operational AI. I'm going to start posting regularly on topics that will help businesses better deploy AI across many use cases and contexts.
Cutting Through the Hype
One of my goals for this newsletter is to cut through the marketing hype and splashy demos that dominate social media and the news every day.
There is an overwhelming amount of information to process. New models, new products, new papers, and companies touting their AI adoption wins. On one side we have hype peddlers that tell us “it’s so over” and AGI will be here soon. On the other side, skeptics like to point to any failed example as a sign that AI isn’t ready for real-world use cases yet.
The truth is somewhere in between.
AI is a transformational technology, but it will behave like all the other waves that came before it. New techniques and capabilities will show up quickly, products leveraging them will come soon after, but broad adoption will crawl behind both. Budgets, compliance, quarterly planning, and inertia hold back deployments, just like they always have.
The public benchmarks aren’t helping companies find the signal in the noise. Many are built using datasets that don't represent what engineers are seeing in the real world. They boil results down to a single headline score and it’s hard to know which specific tasks and use cases the models handle well.
We can do better. We plan to test different models, techniques, and tools on the use cases and messy data that matter for production deployments. Then we’ll share those results in plain language and make them as actionable as possible.
The New Software Capabilities
AI hasn’t made software magical. It has added 10 new capabilities we simply didn’t have three years ago.
Each capability can dramatically improve systems you already use or enable entirely new products, services, and automations that previously weren't possible.
In the next issues we’ll start to unpack these capabilities, show where they work, where they break, and how to deploy them in real projects.
Natural-Language Understanding. Reads messy human words and turns them into tidy facts a computer can understand and use. Can be used to route support emails, pull information from contracts, and understand the exact sentiment in customer reviews.
Reasoning. Connects facts, dates, and numbers and then draws fresh conclusions from them. Flags rule-breaking purchase orders, makes in-the-weeds operational decisions, check loan compliance, and update pricing when costs change.
Planning. Breaks a big goal into clear, ordered steps a system can follow. Can be used to build a month-long marketing calendar, draft a DevOps runbook, and map a day-by-day family trip.
Instruction Following. Sticks to detailed, multi-step directions without drifting off course. Can be used to run KYC checks, draft first-pass legal briefs, and fill in finance reports.
Writing & Summarization. Expands bullet notes into polished pages or shrinks thick reports into quick briefs. Can be used to craft board decks, turn clinic notes into patient letters, and make podcast show notes.
Tool Use (APIs & Plugins). Calls outside apps during a request to finish work. Can be used to update CRM records, send calendar invites, and close Zendesk tickets.
Browser & Desktop Automation. Clicks through on-screen interfaces when no API exists. Can be used to enter data into legacy ERPs, print bulk shipping labels, and upload insurance claims.
Code Generation. Writes clean code on demand in many languages. Can be used to produce helper functions, complex SQL queries, and small microservices with unit tests.
Image & Video Understanding. Reads pictures and video frames the way it reads text. Can be used to find defects on a factory line, extract images from slide decks, and flag risky user content.
Voice Synthesis & Recognition. Listens and speaks in real time with natural cadence. Can be used to draft call-center summaries, translate meetings, and power conversational voice bots.
Coming Up Next
In the coming posts you can expect:
Deep dives into the capabilities above with real world evals, use cases, and practical implementation guides.
Reviews of tools, models, and techniques to help you decide what to implement and how.
Guidance on how to deploy AI within your organization, using insights from people that have done it successfully.
If you have feedback, ideas, or just want to say hello, feel free to respond to this post.
Great stuff Daniel. excited to dive in further.