0:08
[AINews] The Field Guide to Fable
While we congratulate (friend of the show!) General Intuition on their new model and (friend of the show!) Shunyu Yao on their new model, and the world awaits the release of GPT-5.6 Sol Ultra, people are racing to find the limits of Fable 5 before the subscription subsidy ends tomorrow.
Thariq had been working on a “Field Guide to Fable” blog series, and happened to have a keynote planned the day of the relaunch, so he kindly pivoted the entire keynote in one night to give the most timely advice he had, which was released today:
The 4 segments are (my watchalong commentary in italics):
0:00 Introduction and setting the stage for Fable
2:32 Unhobbling Claude: Understanding model behavior
The constraints on a model are often imposed by US - “the harness we put them in, and the way we prompt them”. Therefore when we encounter a new class of model, we should expect to remove or change those harnesses and prompts in order to elicit new behaviors that you otherwise would never see because you were overly limiting (aka hobbling) the model.
Case in point: most people have come to agree with Thariq on the unreasonable effectiveness of HTML.
9:08 Finding your unknowns: Navigating the gap between map and territory
a close cousin to “unhobbling” - if unhobbling is about clearing out outdated knowns, then this is about finding things you didn’t even know you didn’t know.
easiest techniques:
telling claude to do a “blindspot pass” for your unknowns
brainstorm for “wildly different design directions”
interview me - similar to /grill-me, but prioritizing high impact questions
use references: in the case of migrations
keep implementation-notes.md: a running log of underspecified decisions made on your behalf
quiz me - ensure MY understanding
14:29 Dealing with Grief: Reflecting on the emotional shift in coding productivity
What you used to spend weeks on is now done in hours
16:30 Being unreasonable: Demanding good, fast, and cheap results
“Tradeoffs are not real” - because Fable is more capable, you can be more ambitious and not accept tradeoffs.
“Building is easy, generating value is still hard”.
Overall, an excellent talk that we will be mapping out the implications of as the world acclimatizes to the first Fable-class models.
AI News for 7/04/2026-7/06/2026. We checked 12 subreddits, 544 Twitters and no further Discords. AINews’ website lets you search all past issues. As a reminder, AINews is now a section of Latent Space. You can opt in/out of email frequencies!
Tencent Hunyuan’s Hy3 Release and the Open-Weight Frontier
Hy3 lands as a serious open model: Tencent released Hy3 under Apache 2.0, a 295B MoE with 21B active parameters, 192 experts / top-8 routing, GQA, 256K context, and a 3.8B MTP layer for speculative decoding. Multiple posts framed it as competitive with much larger systems on reasoning, coding, and agentic tasks, with particular emphasis on reliability improvements like tool-calling stability and anti-hallucination work @eliebakouch, @HuggingPapers, @ShunyuYao12.
Inference support was unusually day-0 mature: @vllm_project said Hy3 runs natively in vLLM from launch with tool-call and reasoning parsers, MTP speculative decoding, and validated support on NVIDIA and AMD. A follow-up detailed Tencent production kernels now upstreamed into vLLM main, including load-balanced decode scheduling and fused FP8 MoE serving, with reported gains of up to 2.95x on mixed-length decode and latency reductions of roughly 24% TTFT and 17% TPOT versus default backends @vllm_project. Community reaction was strong enough that @Teknium quickly made Hy3 free on Nous Portal for two weeks.
Broader open-model context: Hy3 was immediately compared against GLM-5.2, with some posters arguing Tencent has now joined the very top tier of open-source labs if the benchmark and vibe-test results hold @teortaxesTex, while others still maintained GLM-5.2 as the best currently usable open-weight model in practice @tinygrad, @mbusigin. The net takeaway: the open frontier is compressing fast, and the competition is increasingly about deployment robustness rather than just raw leaderboard deltas.
Agent Benchmarks, Harnesses, and Long-Running Memory
AutomationBench-AA adds a more realistic agent eval: @ArtificialAnlys launched an independent leaderboard for Zapier’s AutomationBench, evaluating agents across 657 tasks and 40 simulated SaaS apps with both objectives and guardrails. Claude Fable 5 led at 48.6%, narrowly ahead of Opus 4.8 at 48.5%, with Gemini 3.5 Flash at 42.6% and GPT-5.5 xhigh at 42.1%. More interesting than the ranking: every model still breaks business rules, and Gemini looked notably strong on objective-per-guardrail-violation and cost efficiency.
4:00
The Six AGaaS Moats
Every business model transition redefines what defensibility means. SaaS had its own moat vocabulary — seat lock-in, network effects, switching costs on stored data, ecosystem depth. Those moats worked because SaaS sold access to tools that humans operate, and defensibility grew around human usage patterns.
AGaaS breaks that vocabulary because AGaaS sells something structurally different. AGaaS — Agentic-as-a-Service — charges for the execution of outcomes by agents, not for access to tools operated by humans.
That single reframing is the disruption. It rewires what gets sold (outcomes, not access), who does the buying (business budgets, not IT budgets), where margin lives (substrate, not interface), and which metrics still predict the future (consumption, not seats). Every SaaS moat was built for a world where the seat was the atomic billing unit. In AGaaS, the seat is gone. The agent is the atomic unit, and the moats have to be rebuilt from scratch around it.
Three sequential inversions define the transition, and understanding them is prerequisite to understanding what defensibility looks like inside the shift:
The operator inversion. The agent, not the human, operates the workflow. The billing unit stops tracking headcount and starts tracking agent activity. Most enterprise software companies have completed this inversion architecturally by mid-2026, even if they have not repriced.
The buyer inversion. The person authorizing the purchase is the executive who owns the outcome (the CRO, the COO, the general counsel), not the IT function that owns the tool stack. Procurement, budget line-items, and the sales motion all rearrange. Most vendors are still mid-flight on this one.
The margin inversion. Gross margin structure shifts from “software plus support” (high fixed margin, low variable cost) to “consumption minus inference” (variable margin dominated by token cost). The P&L reshapes. This inversion is deferred at most incumbents through mechanisms like internal absorption (”Customer Zero”), buybacks, and reclassification, but it becomes visible once consumption revenue scales past a threshold.
The AGaaS transition is what these three inversions look like when they run through the enterprise software category. It has a canonical form, and its canonical form has been mapped in the prior work of this analytical arc:
Follow me on The AI Supercycle as well!
“From SaaS to AGaaS: The Full Cascade” (March 2026) mapped the underlying value-layer inversion — value migrating from the interface layer, where SaaS captured it, down to the substrate layer, where AGaaS captures it — and the five-layer cascade this inversion produces across the enterprise software stack. It also established the three canonical AGaaS pricing models: outcome-based, consumption-based, and hybrid.
“The Four Frictions of AGaaS” (April 2026) mapped the frictions that slow the transition — financial, operational, competitive, commercial — and why they multiply rather than add.
“The Tell: Salesforce Q1 FY27” (May 2026) used the largest pure enterprise-software incumbent as an instrument to locate where the transition actually sits. The verdict: architecturally migrating, mechanically still SaaS. Operator inverted, buyer not, margin deferred.
“The Four Intelligence Moats” (June 2026) — the direct parent to this piece — mapped how the entire AI stack builds defensibility across paradigms: corpus, verifier, harness, container. That piece established the machine-side moat map.
This piece completes the arc by mapping defensibility inside the transition itself. If old SaaS moats no longer apply and machine-side moats alone are not sufficient, what moats does an AGaaS company actually build? The answer is six — three inherited from the machine-side (verifier, harness, container), three new to the buyer relationship (trust, integration, feedback). Together they define what AGaaS defensibility looks like once the transition matures. They also determine, mechanically, which of the three AGaaS pricing models a company can actually charge — a link the closing section of this piece will make explicit.
The order in which the six moats are built matters more than the number of them. They are not parallel. They are serial. Each one unlocks the next. A company that tries to build them out of order finds itself with moats that do not compound — and, more consequentially, with pricing power that stalls at whichever inversion its missing moat was gating.
7:41
Introducing: Misery Marches
This Summer I will be leading the first “Misery March” in the Greater Toronto/Ontario area (TBA). Join the Anarchonomicon Discord Server.
Toronto, for those of you who have never been, is one of the most gorgeous, walkable, safe, pleasant dense urban cities in the World and North America in particular. The old “City of Toronto” or “Old Toronto”, is an amazing old Victorian grid city comprised of almost a hundred villages, town and neighborhoods, that all slowly merged together but all kept their individual quirks… One is even technically an independent republic of 200 people, that the federal government accidentally recognized its independence when their meme-protest became a feel good story 50 years ago (they still pay taxes).
The whole old city, is 16 miles wide and is sort-of a blend of New York and London… but without the Knife, Gun, or “youth” crime. And many lovely hiking trails that wind throughout the city’s hundred+ mile ravine network.
Pleasant summer hikes and walks and groups to walk, abound… You can find gorgeous trails and people to hike those trails or city tours anywhere… And Margaret and her friends aren’t forming a cohesive high trust super fit unit off of it.
Instead we’re going to take subways, streetcars, and buses populated by the strangest smelliest Star Wars Cantina creatures away from that…
We’re going outside Old Toronto, outside Metro Toronto, outside the leafy lovely expanses of endless walkability… And into the nightmare of the unwalkable, concrete MegaCity hellscape sprawl beyond Etobicoke, Scarborough, and the 401, where they shove the airports, and the industry, and the housing projects for all the recent immigrants, where the street signs themselves have been known to stop using the Latin alphabet.
The dead lands beyond Willowdale that RUSH described in The Necromancer.
In high summer. In Ontario humidity. On concrete heat islands that will cook an egg. And we’re going to walk 30km (20 miles). 6-9 hours in that scorching concrete hell where legend tells of second generation old Indian men who have never seen a white face.
Drive one hour outside Vancouver or Montreal, and you’re in the Mountains. Drive one hour outside Toronto… And you’re in Toronto.
We’re going into the heart of darkness, at the worst most miserable time of year to do so…Into lands often described as “Mumbai” or “New New Delphi” or “Bigger Hotter Syria”, Where all the worst trends of Sub-urbanism, mass migration, urbanism, boomerism, sprawl, and post-war mistakes all intersect into Barad-dûr like eyes of pure blighted suburban madness such as “Square One” or “Brampton Shoppers World” or “Woodbine Mall & Fantasy Fair” or “Vaughn Mills” or “First Markham Place” or “The University of Toronto- Scarborough” or “The Pickering Casino Resort”
There’s actually a non-zero chance one of us could die or be driven to madness of heat stroke, violence, mistrusting a corner-store, the parasitic macro-millipedes, or other complications.
20 miles, in the Mid-day sun. Rain or Shine. Back to Civilization and the leafy-green bubble of inner Toronto… The hard way.
But why?
The Rules of a misery march are simple. You and your comrades use google maps to scout out and hunt down the most miserable, blighted, sprawled out, industrial parked, almost unreachable segment of your metro area. The part no one you have ever met has ever been… Then using Public transit, skateboards, uber, taxi, your mom… Etc. To get there. Then you walk back 10-20 miles to the city core or whatever you consider civilization across 6-9 hours.
Depending on the Metro-area and local laws, you may want to be armed… But for Toronto purposes this is unnecessary (Toronto’s Lost Shades and Silent Hill-Esque monsters are nothing if not cowardly).
If anyone has to tap out part way for heatstroke, blisters, or for fear that the sadness will consume them… or their wife nags them to get home after a few hours, we don’t hold it against them. They did the important thing.
10:57
AI: A post-mortem on 'The Blip 2.0' at Anthropic. AI-RTZ #1140
It’s now officially three weeks since the lifting of Anthropic’s June 12 ‘Blip 2.0’ last week by the US Government. Named as such by me after ‘the Blip 1.0’ three years ago, when OpenAI’s non-profit board fired and re-hired founder/CEO Sam Altman over a globally stressful weekend.
Both times an interruption in the AI Tech Wave that was startlingly searing. With lots of unintended consequences still being felt today from the Blip 1.0.
And as I discussed last week, this second one is not yet completely behind us. Especially with the US government still gating both Anthropic and OpenAI on its latest next-generations of frontier ai models. Specifically Mythos from Anthropic and GPT 5.6 from OpenAI.
So it’s helpful to see what the latest post-mortem analysis on how it all went down reveals.
Axios goes through it all in “How the world’s top AI models were revived”:
“The fight that scrubbed the world’s most powerful AI models from the internet featured personality clashes, industry confusion and international backlash.”
“Why it matters: Anthropic’s models are back online, but the impact of its 20-day showdown with the Trump administration will be long lasting.”
“Behind the scenes: It began when Amazon, Anthropic’s partner and investor, sounded an alarm that was later disputed by cybersecurity experts.”
“It warned about a “jailbreaking” issue it found with the AI lab’s latest models, Mythos and Fable — meaning a technical flaw that could have caused a failure of their guardrails.”
“Amazon flagged its concerns to the administration, triggering sweeping export controls. A U.S. official said the government conducted its own tests once it became apparent that the issue needed to be addressed.”
“Cybersecurity experts, however, later wrote in an open letter to the administration that other leading AI models have the same issue Amazon warned about with Anthropic.”
Then the Friday evening events that set if all off with a call from Lutnick:
“On June 12, Commerce Secretary Howard Lutnick, at the direction of President Trump, called Anthropic CEO Dario Amodei.”
“Lutnick made clear to Amodei the issue needed to be resolved fast and alerted the CEO that the company would be receiving a letter imposing sweeping export controls, the U.S. official said.”
“Amodei called Lutnick back that night after receiving the letter, realizing it effectively meant the models would have to be taken offline — to which Lutnick responded that was indeed the goal.”
So clarity at last at that point on the near-term objective. Regardless of broader consequences for the US ‘AI Race’ vs China.
“That decision led to a three-week, multi-agency crash course in AI safety.”
“Anthropic deployed engineers to Washington D.C. According to a U.S. official, the company wanted to prove everything was already resolved and further changes were being fine tuned.”
“But the federal Center for AI Standards and Innovation and the National Security Agency said those changes weren’t good enough, prompting further fixes, according to the U.S. official.”
“Gradually, various agency heads approved of the changes, and on July 1 the models were released, the official said.”
Gears of the Government hard at work indeed.
“Out of all of the administration officials Amazon’s Andy Jassy could have called, it was Treasury Secretary Scott Bessent who first heard about the jailbreaking issue found in the company report, according to a separate source familiar.”
“Bessent was early to sound the alarm on Mythos, work with White House chief of staff Susie Wiles to reengage the embattled company and help get a cybersecurity executive order across the finish line.”
“While technical discussions to address the jailbreaking issue took place in D.C., it was Bessent who stood next to Trump during the G7, where allies called for global cooperation on safety standards.”
Nature abhores a vacuum as they say. And the departure of ‘AI and Crypto Czar’ David Sacks in recent days, left Secretary Lutnick at the AI helm, steering the US ship.
“At the center of the showdown was Lutnick, who also flanked Trump at the G7 meeting while his department’s teams led technical discussions.”
“National cyber director Sean Cairncross, the White House Office of Science and Technology Policy, Treasury Department chief information officer Sam Corcos and the NSA also all participated in technical discussions, according to various sources.”
“Washington mobilized faster to hold scores of meetings and pulled in far more agencies than one would expect for a single technical issue, one source said.”
Then it all guilt from there, continuing over to the G7 meeting in Europe, with Anthropic founder/CEO Dario Amodei in attendance. And meeting briefly with President Trump.
“The tension spiraled amid personality clashes and poor communication.”
“Anthropic eventually understood that in order to be successful, it needed to be on the same side as the government, the U.S.
15:01
What If We Walked Away?
*Taptaptap* Hi, hello, is this thing on?
It became apparent as I fumbled through the settings on my screen that I’ve stepped away from my laptop for quite some time. This summer was meant to be a brain break, an overdue reprieve, a chance to clear my head and, well, step back from social media and reassess what’s most important.
15:22
Why We Foster, a Backstory
This week, we brought home a new foster dog! Meet Indie. She’s an adorable, goofy, curious very-good-girl from the Chicago-based rescue we regularly work with, One Tail at a Time. Coming in at 70+ lbs, she’s also the largest dog we’ve ever invited into our home. Indie is all legs with the most kissable face, and I often jump-scare myself when I catch her out of the corner of my eye, thinking she’s a whole dang human walking into the room.
Indie is the 5th pup we’ve fostered through One Tail, and she won’t be the last. I already know she’s going to make her future family very, very happy.
15:55
My Experience Starting HRT
I imagine you’ll be post-menopausal before you’re 50.
I had just finished going over my symptoms and backstory with my hormone doctor. I let her know that, physically, I feel great! The best I’ve felt in my entire adult life.
I told her about the shifts I noticed in my health starting in my mid-30s.
16:14
A Transition to Fall: De-glorifying the Idea of Busyness
The last few posts have been pretty heavy topics, so I’d like to lighten things up this time! I always feel a wave of relief on the first day of fall, like, okay, we can officially snuggle into our sofas, guilt-free. It feels like permission to settle into a more quiet routine and whipping up heartier, family favorite recipes. And, most importantly, let the Halloween costume brainstorming commence!
It’s the time of year where we stick to a 9-5 workday, help Lucy with homework while one of us cooks dinner, and we end the day with pajamas and an hour of coloring, crafts or something creative.
16:49
Do You Decorate Your Home for Halloween?
Aside from the college day Halloween parties where we’d dress to the nines in our pulled-together thrifted costumes (only to consume foamy beer from kegs while house-hopping and trying to grab the attention of a crush) — er, I digress — we’ve never been ones to lean into this holiday. And it wasn’t until Lucy was 3 that she began questioning why other houses were covered in spiderwebs and inflatable ghosts.
The following year, she helped us throw synthetic cobwebs onto our front gate; this is the same year I learned I despised the stretchy, messy fluff.
17:22
This Is Not the Chicago I Know
This is not about any Chicago you’ve seen in the news lately.
It’s not about middle-of-the-night raids on apartment buildings, clouds of tear gas thrown by ICE at the end of my street, or racial profiling paired with lack of due process.
It’s not about the careless detaining of my ward’s Alderperson, the horrific intimidation of masked and armed officials roaming my neighborhood playgrounds, grocery stores and hospitals, or the inhumane way I see our neighbors being torn from their children.
This is not about the withholding of funds for Chicago public schools out of spite, and the emails I fr
