Independent Technologist | Global B2B Thought Leader & Influencer | Advancing Human-Centered AI & Digital Transformation
Global
- Tweets 53,903
- Following 64,752
- Followers 351,337
- Likes 266,445
You and your AI do not need to split the work 50/50; what matters is the right contribution at the right moment, not an equal share.
Good collaboration is not measured by symmetry, but by whether each side contributes what the task needs when it needs it.
Microblog @antgrasso
In practice, the balance can shift from one task to the next.
A procurement team may let AI scan thousands of supplier records and flag unusual changes. Most of the volume sits with the machine.
Then one case reaches a buyer who knows that a supplier is going through a temporary production change. Context changes the interpretation, and one human judgment can outweigh hours of machine processing.
AI may then return to the workflow and handle the next batch.
No symmetry required.
Responsibility follows a different rule. If an AI-supported recommendation affects a supplier decision, the organization still owns the outcome.
So do not measure collaboration by how much work each side does. Ask whether AI is handling the work it can do well and whether people enter where judgment is needed.
A 50/50 split looks neat on a slide.
Real collaboration follows what the work requires.
#HumanAICollaboration #FutureOfWork
6
37
80
1,585
Adding more agents is easy. Coordinating their work is the harder problem.
The shift is letting specialists work in parallel, share context, and bring people in when judgment or approval is needed.
That feels much closer to real teamwork.
@hyperagentapp #HyperagentPartner
Your agents have entered the group chat
We just launched Rooms: a multiplayer space for agents and humans to work together
In a Room:
> many specialized agents execute in parallel
> your entire human team can message the same thread
> agents surface approval and review requests when needed
Hire your team of agents on Hyperagent
Paid partnership (ad)
8
45
104
4,246
Cloud scalability means adjusting IT resources as demand changes, adding capacity when needed and reducing it when demand falls.
Companies avoid waste, keep services running during busy periods, and adapt faster without investing upfront.
Microblog @antgrasso
5
51
95
2,771
Antonio Grasso retweeted
Deel links adding >$140M in ARR over 90 days without increasing headcount to its internal use of Akai.
What interests me is how Akai learns the path through a workflow, including edge cases, and turns it into something teams can reuse and adapt.
EXCITED TO LAUNCH: Akai (akai.run)
Deel added >$140M ARR in 90 days without increasing headcount by automating~600 Full Time Employees' equivalent in work with Akai.
Akai was an internal tool to automate our painfully repetitive operations in Finance, HR, Accounts Payable, and Compliance, etc. We never intended to make this a product.
But we watched revenue per employee grow from $130K to $215K
We built >8k agents that do the work of ~600 employees
It had such a dramatic impact on our business that today we are launching it for everyone.
How it works: Say you're automating payment reconciliation:
1. Record your screen while manually matching a messy transaction and Akai will capture your screen, voice, server requests
2. Akai will see that you pulled unformatted wire transfer info from an archaic bank portal, put it in some excel sheet, checked NetSuite invoices, payment history, and put a ticket on Zendesk
3. Akai reads between the lines and build a workflow + steps + conditional guardrails. It learns tacit edge cases, like resolving malformed invoice references without you writing a single regex
4. Simply connect NetSuite, your ledger, Zendesk, PSPs, and even legacy bank portals with zero API access
5. Run the workflow and tell it what to adjust in plain English: "strip slashes on wire memos and auto-apply partial payments." It adapts instantly
6. Once it works for you, add 100s of colleagues. Your entire payment ops team forks and extends the workflow for new PSPs, secondary ledgers, or regional settlement rules
7. We automated 85% of our payment reconciliation end to end, eliminating 500+ hours of soul-crushing manual grunt work every single week.
Claude Code/Codex can't do this in multiplayer mode. Every person rebuilds the same skill from scratch in their own way.
Deel built Akai to:
1. Understand backend operations edge cases (it had to work for our 7000 person team first)
2. Collaborative across 1000s of employees
3. Self-Learning from millions of runs
4. Optimises cost and gets cheaper every run
We're so confident that we're announcing an Automation Guarantee:
If our engineers can't automate a thousand of hours of work in your first 30 days, you get a full refund.
Book a demo: akai.run if you're an exec at a company with hundreds of employees
Paid partnership (ad)
3
45
66
4,570
You have heard that AI consumes energy. Fine. But what exactly is using that energy while you type a prompt and wait for the answer?
What you see on screen is only the result, so the energy behind it goes beyond the few seconds spent generating the answer.
Microblog @antgrasso
There is no single energy cost hiding inside an AI answer.
Some electricity is already being used before your prompt arrives. The infrastructure may be powered and ready to serve requests.
Then you press Enter.
The model has to process your prompt before it can answer. Longer inputs, especially when they include a lot of context, can require more computation.
Then comes generation. The model produces the response step by step, using computation as the output grows. Longer responses generally require more work.
The model is not working alone. Memory and networking keep data moving through the system, while cooling uses electricity as the hardware runs.
Which part uses the most energy? There is no fixed answer.
It depends on the model and the workload. Model size, prompt and response length, modality, and infrastructure efficiency can all change the total.
That is why saying “one AI prompt consumes X” can hide more than it explains.
The energy is in the whole slice.
Like a cheesecake, you enjoy the result without necessarily thinking about what went into every layer.
#ArtificialIntelligence #EnergyEfficiency
3
61
114
3,296
Antonio Grasso retweeted
AI can still amaze you, but remember that beneath every fluent answer there is software doing a great deal of mathematics.
Words become numbers, and specialized processors keep performing the calculations that make the response appear effortless to you.
Microblog @antgrasso
Ask an AI model a question.
Your words are converted into numerical representations. During inference, the model performs vast numbers of matrix and tensor operations to estimate what should come next.
The fluency appears on the screen.
The computation stays underneath.
A general-purpose CPU can run AI workloads, but large-scale AI benefits from processors designed for parallel computation.
Graphics processing units, or GPUs, and tensor processing units, or TPUs, can execute many of these mathematical operations at the same time.
That difference helps explain why specialized compute has become such an important part of AI infrastructure.
None of this makes AI less impressive.
Quite the opposite.
Understanding what happens underneath helps us appreciate the engineering without imagining magic where there is computation.
AI can still amaze us.
Awareness simply lets us admire it with our eyes open.
#AI #AIComputing
6
69
113
3,598
If you are making room on your desk for a quantum computer, stop. You do not need one there: quantum computing is a specialist, not a faster PC.
Use it when a problem may benefit from a different way of computing, not just because you want more speed.
Microblog by @antgrasso
Most of the work will still be done by classical computers.
Take medical research.
A classical system can manage the data and the workflow, while a quantum processor is assigned a molecular simulation that may benefit from quantum computation. The result then returns to the classical workflow.
Hybrid approaches like this are already being explored in drug discovery and biological research.
Weather research gives another example.
Forecasting still depends on classical computing, but researchers are testing quantum-classical methods for parts of the process such as data assimilation. The quantum processor tackles a specific task; it does not replace the forecasting system around it.
In a hybrid system, classical computing keeps doing what it already does well. Quantum computation enters where its different way of computing may offer an advantage.
So the question is not: when will a quantum computer replace the machine on my desk?
A better question is: which part of a problem might justify calling in a quantum specialist?
Quantum computing is not the next PC.
It is another kind of compute for specific jobs.
#QuantumComputing #HybridComputing
9
60
108
3,304
Antonio Grasso retweeted
AI agent is becoming the label of choice for almost any AI workflow. But autonomy is more than a name on a product page.
Marketing can stretch the term, so before calling something an agent, look beyond the label and ask what the system can really do.
Microblog by @antgrasso
Forget the label for a moment. Look at what the system does.
If every step is predetermined and the software follows the sequence, that is automation, even if AI is inside it.
Agency begins when the system can pursue a goal and use what happened to choose what to do next.
The agent is not the model. It uses a model for reasoning. The surrounding system keeps track of the goal and current state, then gives the model access to tools when action is needed.
For organizations, that distinction matters.
A system that can choose its next action needs a mandate and boundaries. Accountability still belongs to people.
So before buying the word agent, ask a more useful question: can the system change what it does next because of what just happened?
Marketing can rename a workflow. Agency has to appear in the behavior.
#AIAgents #AgenticAI
20
69
109
3,036
Antonio Grasso retweeted
Polkadot connects independent blockchains through shared security, allowing data and assets to move across chains while transactions run in parallel.
DOT supports staking, governance, and network operations.
Microblog @antgrasso
#Polkadot #CryptoExplained
11
97
306
7,259
Antonio Grasso retweeted
I know Activation Steering sounds complicated, but the idea is simple: adjust AI behavior at inference without fine-tuning or retraining the model.
That can save time and cost when changing model weights is unnecessary for the behavior you want to change.
Microblog @antgrasso
That can save time and cost when changing the model weights is unnecessary.
An internal AI assistant, for example, may keep producing answers that are longer than the team wants. Instead of creating another tuned version of the model, researchers can test whether steering selected activations during inference moves the responses toward the desired behavior.
Another case is a model that tends to agree too readily with what users tell it. Activation steering can be tested to influence that tendency while the model is generating its response.
The base weights remain unchanged.
Remove the steering intervention and you are still working with the same underlying model.
Sounds convenient? It can be.
But the effect is not guaranteed to behave the same way across models or tasks. A change aimed at one behavior can also influence something you did not intend to modify.
So the business case is more specific than “no more fine-tuning.” For targeted experiments or selected behavioral adjustments, activation steering may offer a faster and less expensive route before committing to a new tuned model.
Promising technology, yes.
A shortcut around testing, no.
#ActivationSteering #AIEngineering
4
61
100
3,385
Antonio Grasso retweeted
A text carries a thought; it is not the thought itself. Until software can identify who supplied the thinking, AI detectors can't prove authorship.
Calling a text 100% AI-written from patterns alone is no proof at all. It is confidence dressed as certainty.
Microblog @antgrasso
Take this post.
I am Italian, and I think more naturally in Italian. I develop the idea and connect the concepts in my own language, then I explain to AI what I want to say.
I do not want my thinking to come out in schoolbook English. I want to express it in the professional English that fits who I am and the work I do.
So I talk to AI. I challenge words, change sentences, reject what does not sound like me, and keep working until the text says what I intended to say.
Then someone takes the finished post, puts it into an AI detector, and the software reports: “100% AI-written.”
Really?
The detector sees the final language. It does not see who had the idea or who directed the reasoning that produced it.
And this is where the problem gets worse for non-native English speakers. Their writing patterns can already be misclassified by AI detectors, even when the text was written by a human.
So if I think, connect, decide, and then use AI to express that thought better in English, does the machine suddenly own the authorship?
Ma fatemi il piacere.
Use AI detection as a signal if you need one. Do not use it as proof of who produced the thinking behind the text.
A detector can inspect words, but it cannot inspect authorship.
#AIDetection #ResponsibleAI
3
68
99
3,855
Antonio Grasso retweeted
“Festina lente.” Make haste slowly.
Digital tools can speed up work, but transformation begins by questioning the system they are asked to accelerate.
Before adding speed, ask whether the process should change first and whether its logic makes sense.
Microblog by @antgrasso
Here is the trap.
Move a five-step approval process into software and celebrate because everything runs faster.
The workflow is faster. The logic is unchanged.
If two approvals add no judgment, remove them. If employees keep entering information that already exists elsewhere, connect the systems instead.
Only then does technology have something better to accelerate.
A redesigned workflow works when the people using it understand why decisions move differently and can operate within that logic.
Technology can execute an existing process faster, but it cannot decide whether that process deserves to exist in the same form.
Before asking technology for more speed, ask the system whether it needs a different direction.
#DigitalTransformation #ProcessRedesign
4
67
109
4,356
Antonio Grasso retweeted
Attention is the scarce resource in entertainment.
Pocket FM built its model around hours when screens are impractical and audio still fits into the day.
Sherpa adds another layer by helping writers turn a premise into serialized fiction at scale.
Introducing Sherpa: the most advanced fiction writing AI
We accelerated from $250M in ARR to $500M because Sherpa helped increase content production by 1200% in 1 year
Sherpa was trained on 5.5B hours of playtime with minute by minute dynamic retention data.
550K+ creators have produced 2.6M hours of content annualised using it
Pocket FM is like Netflix for audio-only dramas, with our own pool of one-person studios.
10% of eligible writers on Pocket FM make >$200K
One blockbuster produced >$100M in revenue
3 writers have become millionaires in <2 yrs
We built Sherpa to enable anyone to make >$1M by writing world-class fiction stories:
1. The Idea: Drop a 1-2 sentence concept. Sherpa interrogates it like a veteran editor on tension, stakes, and psychology
2. World & Characters: It builds out the complete lore, tone, and character psychologies
3. Sub-Plot planning: Breaks the premise into arcs, arcs into episodes, and episodes into scenes
4. Scene-by-Scene Generation: Outlines and drafts entire episodes, with you able to steer, rewrite, or override anytime
5. Editorial Review: Stress-tests every draft for pacing, engagement drop-offs, prose, and coherence before it locks
6. One-Tap Production: Pick a voice, convert to audio drama, and publish directly to Pocket FM’s millions of listeners
7. Global Scale & Monetization: Revenue-share on performance, with automatic localization so you earn across international markets
Test Sherpa for free here: pocketfm.com/sherpa
_____________________________________________
Generic LLMs fail at serialized fiction because they lack a long-horizon narrative reward function.
Sherpa solves this through three core technical leaps:
1. Narrative World Model (State Tracking & Retrieval): Context windows degrade over long runs. Sherpa constructs an evolving semantic knowledge graph tracking character states, secrets, and plot dependencies. High-speed retrieval surfaces exact context on demand, maintaining zero continuity decay across hundreds of episodes
2. Hierarchical Story Planner: When writing a 500-episode story like Naruto, you need to plan 100s of sub plots. Rather than generating linearly, Sherpa decomposes narrative across discrete levels: season -> arc -> sequence -> episode -> scene.
Rather than generating everything upfront, like a generic LLM, Sherpa uses progressive planning and dynamic replanning. As the story evolves, it identifies what changed, traces the downstream impact, and replans only the affected parts.
3. Prose Engine (Trained on series' retention data): LLMs write robotically, but serial fiction needs emotion, tension, pacing, and dialogue that sounds like real people.
Sherpa's Prose Engine was designed specifically for storytelling. It was built on 1B+ tokens of Pocket's own stories, trained by learning from what listeners engage with, where they drop off, and what keeps them hooked.
Feedback is taken from specialized evaluator models that measure every scene against a 40-item checklist. (Evaluator models were benchmarked against human reviewers and matched them 80–90% of the time.)
_______________________________________________
Owning distribution and creation puts us in a very unique spot.
More shows -> More data -> Sherpa becomes better -> more creator success -> more creators -> more shows
Pocket FM has already seen one $100M IP. I believe Sherpa will soon lead to dozens of single-person studios creating billion-dollar shows.
Most people are scared of AI but I think it'll unlock more human creativity, help creators earn more, and bring the next great IPs to life. This will create millions of jobs and new income streams.
Paid partnership (ad)
5
53
64
5,327
Blockchain is to crypto what the internet is to email. One made the other famous, but the technology can do much more.
Crypto is one application; blockchain can also help organizations share records and verify information across companies without one owner.
Microblog @antgrasso
Forget crypto for a moment.
Imagine three companies working on the same supply chain. Each keeps its own records, so they spend time checking which version is correct.
With blockchain, they can share a record of what happened. Each update follows agreed rules, and previous entries are difficult to change without the others noticing.
A product can carry a shared history as it moves between companies.
A digital credential can be issued by one organization and verified by another.
A smart contract can trigger an action when a defined condition is met.
Crypto uses the same technology to record and transfer digital value. It is one application, not the definition of blockchain.
The question is: do several parties need to share and verify the same record without giving one of them control?
If yes, blockchain may be worth considering.
#Blockchain #DigitalTrust
6
69
116
5,161
Antonio Grasso retweeted
Shared AI context should not mean shared access.
mio brings company memory into Slack while preserving permissions. The same AI employee can serve the whole team without giving everyone the same access.
That boundary belongs in the architecture from day one.
#EnterpriseAI
Today, we’re killing the AI assistant
Introducing mio.xyz, the first AI EMPLOYEE 𝘆𝗼𝘂𝗿 𝘄𝗵𝗼𝗹𝗲 𝘁𝗲𝗮𝗺 𝘀𝗵𝗮𝗿𝗲𝘀
In beta since July, Mio saved teams 1,000s of hours and completed 10,000s of tasks
You don't need another agent. You need a 𝘀𝗵𝗮𝗿𝗲𝗱 one with your team
Hire Mio today, $100 free credits for the first 100 companies
Paid partnership (ad)
16
54
120
5,211
Antonio Grasso retweeted
Seven months, $50M in annualized revenue. The stronger signal is what companies are paying for: an AI employee who does useful work inside teams and gives people time back for judgment and higher-value work.
@viktor_com
Paid Partnership.
Exciting update!
We've hit $50M in annualized revenue. After just 7 months!
BUT we're not even 0.002% into our mission.
We're just getting started.
Paid partnership (ad)
7
52
97
5,087